All servers / Identity / Didit MCP server
Didit MCP server
Didit MCP Server - AI agent integration for identity verification APIs
Tested by ConnectATDScore 88 out of 100. Tested 2026-11-18 on package version 5.1.73. Package: @didit-protocol/mcp-server · source
Score breakdown
| Runs | 18 / 20 | |
|---|---|---|
| Tool descriptions | 18 / 20 | |
| Package safety | 20 / 20 | |
| Maintenance | 20 / 20 | |
| Adoption | 12 / 20 |
What we saw
- Starts and answers the MCP handshake: yes (2.8 seconds including download)
- Tools offered: 176
- Last published: 2026-11-18
- Weekly downloads (npm): 1,906
- License: MIT · Direct dependencies: 6
Want this server set up for your team?
ConnectATD builds AI automation and wires up MCP servers like this one. Tell us what you are trying to do.
Talk to ConnectATDTools and how they are described
Agents choose tools from these descriptions. Under each name you can see how many of its parameters are explained.
didit_account_register2 of 2 parameters describedRegister a new Didit account. A 6-character verification code is sent to the email. Follow up with didit_account_verify_email.
didit_account_verify_email1 of 2 parameters describedVerify email with the 6-character code. Returns the new organization with its live application and its sandbox_application (each with client_id and api_key) and an access_token, and signs the local server in: the tools that follow are authenticated automatically. If the account already has two-factor authentication it returns 2fa_required (and a temp_token when it has an authenticator) instead of tokens: finish with didit_account_verify_2fa. If the account's organization enforces SSO it returns sso_required: no token or API key is issued here, so ask the person to sign in through their organization's SSO in the Didit console in a browser; do not retry.
didit_account_resend_otp1 of 1 parameter describedResend the 6-character email verification code for a pending registration. Codes expire after 10 minutes.
didit_account_login0 of 2 parameters describedLog in to an existing Didit account. Signs the local server in, so the tools that follow are authenticated automatically, and returns access_token. If the account has two-factor authentication it returns 2fa_required and a temp_token instead: finish with didit_account_verify_2fa. If the account's organization enforces SSO it returns sso_required: no token is issued by password, so ask the person to sign in through SSO in the console, or use didit_account_device_login_start. For an account that signs in with a passkey or SSO (no password), use didit_account_device_login_start instead.
didit_account_device_login_startno parametersSign in WITHOUT a password or local browser (OAuth device code): returns a short user_code and a URL the person opens on any device to approve this connection. Works for passkey and SSO accounts. Then call didit_account_device_login_poll. Credentials are stored locally and used automatically afterwards.
didit_account_device_login_pollno parametersWait (up to about 45 seconds) for the person to approve the sign-in started with didit_account_device_login_start. Returns status pending (call again) or connected with the granted scope.
didit_account_verify_2fa2 of 2 parameters describedFinish logging in to an account with two-factor authentication, using the temp_token from didit_account_login and the current 6-digit authenticator code (or an unused backup code). Signs the local server in, like didit_account_login, and returns access_token.
didit_org_list1 of 1 parameter describedList the organizations you belong to (each has an id to pass as organization_id to other tools). In hosted OAuth mode no arguments are needed. In stdio mode the sign-in from login/verify_email is used automatically.
didit_org_list_applications2 of 2 parameters describedList the applications in an organization (each has an id to pass as application_id). In hosted OAuth mode pass only organization_id.
didit_org_get_application1 of 3 parameters describedGet application details (client_id etc.). The api_key is REDACTED — a masked preview + api_key_set flag are returned, never the raw secret.
didit_org_reveal_application_api_key4 of 4 parameters describedReturn the RAW, un-redacted api_key for an application. This exposes a LIVE SECRET — only call when a human explicitly needs the key to integrate; never log or persist it. Requires confirm:true AND an explicit organization_id + application_id (no defaults).
didit_context_getno parametersReturn ALL organizations you can access with their applications nested, plus the default org/app when unambiguous — in ONE call (replaces didit_org_list + per-org didit_org_list_applications). Call this first to discover ids.
didit_session_search10 of 10 parameters describedSearch verification sessions ACROSS ALL your apps and organizations in a single call — the efficient way to answer "the last 5 in-review sessions across my apps". Aggregates server-side and returns newest matches first, each tagged with its organization/application. Omit organization_id/application_id to span everything; pass them to narrow scope.
didit_transaction_search8 of 8 parameters describedSearch transaction-monitoring (AML) transactions ACROSS ALL your apps and organizations in one call. Omit organization_id/application_id to span everything; pass them to narrow. Returns newest first, each tagged with its org/app.
didit_case_search8 of 8 parameters describedSearch case-management cases ACROSS ALL your apps and organizations in one call. Omit organization_id/application_id to span everything; pass them to narrow. Returns newest first, each tagged with its org/app.
didit_vendor_user_search9 of 9 parameters describedFind vendor users (your end customers) ACROSS ALL your apps and organizations in one call — e.g. locate a customer by vendor_data/email/name without knowing which app they're in. Omit organization_id/application_id to span everything; pass them to narrow. Each hit is tagged with its org/app.
didit_vendor_business_search9 of 9 parameters describedFind vendor businesses (KYB companies) ACROSS ALL your apps and organizations in one call. Omit organization_id/application_id to span everything; pass them to narrow. Each hit is tagged with its org/app.
didit_analytics6 of 6 parameters describedAggregate verification analytics ACROSS ALL your apps and organizations in one call - the efficient way to answer questions like "how many people tried phone verification but dropped off in the last 15 days?". Returns summed status counts (request_breakdown), a feature_funnel (how many sessions REACHED each step: OCR, LIVENESS, FACE_MATCH, AML, POA, PHONE, EMAIL, ...), warning_stats by risk code, the previous period's request_breakdown, per-workflow session totals and a recomputed conversion_rate, for a date window. Omit organization_id/application_id to span everything; pass them to narrow.
didit_session_create12 of 12 parameters describedCreate a verification session. Requires workflow_id (created in the Console Workflows page) — the workflow defines which steps run and whether the session is KYC or KYB. Returns session_id, url, and session_token.
didit_session_list11 of 11 parameters describedList verification sessions for ONE app. If you have multiple apps (or pass no scope) this automatically spans every app, newest first - but didit_session_search is the canonical cross-app tool and what to reach for to answer "my last N sessions". HOW MANY: `count` stops at 100 (100 means "100 or more"); when it does, the result adds `total_count` (the exact number, from the console's own tab counts) or `count_is_lower_bound: true`. Answer "how many" from total_count, never report 100 as the number; limit:1 is enough to count.
didit_session_get_decision1 of 1 parameter describedGet the full verification decision and all extracted data for a session.
didit_session_get_location_evidence1 of 1 parameter describedPrecise Location (GEOLOCATION) evidence of ONE session, per Location node: `config` (what judged it, every action resolved through the preset), `result` (the decision's verdict), EVERY attempt incl. retired ones (`retired:true`), `reference_points` (device, IP, address; `area_reading` is the earlier attempt a carried area verdict stands on), `distances`, `areas` (inside / blocked / undetermined area ids, `nearest_allowed_km`, `allowed_result`), `anonymizers` (vpn, proxy, tor, spoofing, emulator; null = cannot tell) and `history`: the same person's earlier checks (same vendor_data) that LOCATION_CONSECUTIVE_MISMATCHES and travel tolerance are judged from. The decision keeps only the accepted attempt: use this for 'which attempt was outside', 'why consecutive mismatches' or 'why was the distance excused'. Holds precise coordinates: quote only what the question needs. `nodes: []` means no Location attempt was recorded; a 404 'Precise Location is not available.' means the feature is off on this deployment.
didit_session_simulate7 of 8 parameters describedSANDBOX ONLY: finish a session of a sandbox application (mode 'sandbox' in didit_context_get) with the outcome you pick, without anyone going through the verification flow, then read it with didit_session_get_decision and its webhooks with didit_session_webhooks. Never billed. Pass new_status ('Approved', 'Declined', 'In Review', 'Expired', 'Abandoned'), a scenario slug ('approve', 'decline_aml_hit', 'decline_document_expired', 'review_aml_possible_match', ...) or a risk. A live session answers not found.
didit_session_update_status6 of 7 parameters describedOverwrite a session's decision by approving, declining, or requesting resubmission; optionally send a notification email that cannot be recalled. Before calling, obtain explicit user confirmation of the session, new decision, and any email recipient. For resubmission, pass new_status='Resubmitted' and nodes_to_resubmit; already-approved steps are kept.
didit_session_update_data6 of 18 parameters describedCorrect the KYC data extracted from the ID document on a session (reviewer override). Only send the fields you want to change.
didit_session_update_poa_data3 of 10 parameters describedCorrect the Proof of Address data extracted from the POA document on a session (reviewer override). Only send the fields you want to change.
didit_session_delete1 of 2 parameters describedPermanently and irreversibly delete a single verification session (KYC or KYB) - its verification data and stored media. Blocklist entries, issued share tokens, already-queued webhook deliveries, consumed credits, and the parent user/business are not affected. Requires confirm:true - only pass it after the user has explicitly confirmed this exact deletion.
didit_session_batch_delete5 of 5 parameters describedPermanently and irreversibly delete multiple KYC sessions by session number, or EVERY KYC session in the application with delete_all:true (KYB sessions are never touched; delete those individually). Requires confirm:true - only pass it after the user has explicitly confirmed this exact deletion.
didit_session_generate_pdf2 of 3 parameters describedReturn download_url, expires_at and expires_in (300 seconds) for the original Didit session PDF. Fetch the URL to save the PDF; no Authorization header is needed. Treat the URL as sensitive. Requires configured MCP HTTP PDF downloads.
didit_session_get_signature_evidence5 of 5 parameters describedReturn download_url, content_type, expires_at and expires_in (300 seconds) for a signer's evidence PDF of an E-signature step: level and format, signer, IP address, each document with its original SHA-256 and field values and the SHA-256 of every signed file, plus, for an Advanced (AES) signature, the verified identity session, the code confirmation, the certificate and the timestamp. Pass the evidence_file path's signer as signature_id; a Business Verification step with several signers needs it, otherwise the route picks one. Fetch the URL to save the PDF; no Authorization header is needed. Treat the URL as sensitive. Requires configured MCP HTTP downloads (see the README).
didit_session_get_signed_document5 of 5 parameters describedReturn download_url, content_type, expires_at and expires_in (300 seconds) for the signed PDF of one document of an E-signature step: the document the signer was shown with their field values stamped and the evidence page (unless signature_evidence_page is off). An Advanced (AES) PDF also carries the PAdES signature with its timestamp; a Simple PDF has no digital signature, seal or timestamp. Didit checks the file against sha256_signed first (409 on a mismatch, nothing returned). Use the id of the document in signatures[].documents[]. Treat the URL as sensitive. Requires configured MCP HTTP downloads.
didit_session_get_signature_field_image5 of 5 parameters describedReturn download_url, content_type (image/png), expires_at and expires_in (300 seconds) for the signature or initials a signer applied to a field. Use the field id from the image_url of signatures[].documents[].fields[]. Treat the URL as sensitive. Requires configured MCP HTTP downloads.
didit_session_validate_signatures5 of 5 parameters describedRe-check the signed documents of an E-signature step and save the result in each signer's validation. Didit re-reads every signed PDF, checks its SHA-256 and, for an Advanced (AES) signature, validates the PAdES signature, the certificate, the revocation data and the timestamp (a Simple signature has no digital signature, so only file integrity is checked). Nothing is signed again. Without signature_id every signer of the step is checked and the call fails with 409 if any has not signed; 400 means a stored signed PDF no longer matches its recorded SHA-256. Returns the checked signers.
didit_session_list_reviews4 of 5 parameters describedList the review history and activity log for a session, newest first. One page: 50 entries unless `limit` says otherwise, with `count` as the total and `offset` to read the ones after.
didit_session_add_review4 of 5 parameters describedAdd a review note to a session's audit trail, optionally changing its status.
didit_session_share1 of 2 parameters describedShare a verified session with a trusted partner for reusable KYC (B2B session sharing).
didit_session_explain_decision1 of 1 parameter describedWHY a verification session ended Approved / Declined / In Review, as a deterministic trace read from the record: the feature that decided it (strictest wins), each warning that had an effect with its raw value, every triggered status rule (field, operator, configured vs actual value), the human override if a reviewer changed the automatic verdict, and for each cause the workflow config key that governs it (`lever`, with its CURRENT value read from the workflow graph; only keys that exist in the feature-config contract; null means it is not a workflow setting). `workflow.graph_read:false` means the graph could not be read, so no current values. Use this for every 'why was it declined / in review / approved', 'it should have been X' and 'how do I make these go to X' question INSTEAD of reading the whole decision — it never returns images, extracted fields or PII. `informational` lists warnings that changed nothing. `unexplained:true` means a decided session (Declined / In Review) whose record shows no effective cause (e.g. minimized age-assurance logs): say so, never invent one. `lifecycle_status:true` (Expired, Abandoned, Not Started, In Progress, Resub Requested…) means nothing decided it — the status is the whole explanation.
didit_session_webhooks2 of 2 parameters describedDelivery log of the webhooks Didit sent for ONE session (last 30 days): destination URL, HTTP response status, timing, signing headers (X-Signature/X-Timestamp) and a capped copy of the payload. The way to check an integration end to end: create a session on a SANDBOX application with sandbox_scenario 'approve' (unbilled, fires status.updated immediately), then call this with wait_seconds so the call waits for the delivery instead of you polling. A 2xx response_status means the customer's endpoint received it; 401/403 usually means their signature check (must be HMAC-SHA256 hex of the RAW body); an empty list after the wait means the endpoint never got the event (not public, or the destination is disabled).
didit_session_webhook_resend2 of 2 parameters describedResend one recorded webhook delivery of a session to the configured endpoint (the session page's own Resend). Use after the customer fixed their endpoint so they do not need a new session; returns the new delivery with its response_status.
didit_session_import_shared1 of 1 parameter describedImport a shared verification session from a partner (Reusable KYC).
didit_session_create_import14 of 14 parameters describedCreate a bulk import job: a hosted CSV/NDJSON file (source_file_url) of historical verifications, or (source=provider) a migration from another identity verification provider. File imports use Didit's canonical schema; label a record's source with the optional per-row `provider` column in the file. A provider migration reads Sumsub, Onfido, Persona and ComplyCube with the customer's READ-ONLY key (`credentials`, named as that provider's credential_fields from didit_session_list_import_providers); Veriff, iDenfy and MetaMap are read by the record IDs in source_file_url, and the others import an export file. The key is secret: ask the user for a read-only key, never log or repeat it, and never reuse it. Didit encrypts it, never returns it and erases it when the import ends. Prefer dry_run:true first: it counts what would import and writes nothing, ends READY, and didit_session_manage_import action=start then imports for real. Record IDs are namespaced by provider, so re-runs and moving between a file and a provider migration never import a record twice. An application runs at most 4 active import jobs at once (a fifth answers 429); a job a deploy interrupted resumes by itself. Imports must be enabled for the organization by Didit and are refused on a sandbox application (403).
didit_session_list_import_providers2 of 2 parameters describedList the 15 providers verifications can be migrated from: what each needs to connect (credential_fields, regions), whether its API can list records, the permissions its read-only key needs and its default status mapping.
didit_session_check_import_provider6 of 6 parameters describedRead a provider with the customer's read-only key before creating a migration. Nothing is stored. For providers read by record IDs, pass record_ids (the first record ID of the export). The key is secret: never log or repeat it.
didit_session_manage_import5 of 5 parameters describedAct on an import job: start (a dry run that ended READY imports for real, with the status_mapping, face_templates and lawful_basis_confirmed sent now), pause, resume, cancel (what was imported stays; the uploaded file and the key are erased) or retry (runs an ended migration again as a NEW job with a new key in credentials, since the old one was erased; everything already in Didit is skipped).
didit_session_get_import_template2 of 2 parameters describedDownload the canonical import template (column headers and format) used by didit_session_create_import.
didit_session_get_import3 of 3 parameters describedGet the status and summary of a bulk import job.
didit_session_get_import_errors3 of 3 parameters describedList per-row errors for a bulk import job (rows that failed validation or processing).
didit_workflow_list5 of 5 parameters describedList verification workflows. With multiple apps (or no scope) it auto-spans every app, each row tagged with its org/app. To find one workflow by id/label across all apps, prefer didit_workflow_search. The `features` on a row do NOT tell you whether a workflow checks age: age assurance done from the DOCUMENT (age restrictions on the OCR step) shows no feature of its own. For any question about which workflows do age assurance, pass `include_age_assurance:true` — it annotates each row with `does_age_assurance`/`methods` server-side, which is the only reliable way to find them. In one application the answer is ONE PAGE, most recently changed first: 20 workflows unless `limit` says otherwise, with `count` as the total and `offset` to read the ones after. Across applications it is the first 50 unless `limit` says otherwise, with `total_matched` as the total and the same `offset`. To count or filter workflows, read until you have them all.
didit_workflow_create7 of 12 parameters describedCreate a SIMPLE (linear) verification workflow from an ordered list of features. Send `features` in execution order (uppercase values; put dependency features first, e.g. OCR before FACE_MATCH/NFC/DATABASE_VALIDATION). The MCP assembles them into a real workflow graph and publishes it; pass status:'draft' to keep it unpublished. A bad feature order fails validation and leaves a fixable draft. A workflow verifies EITHER a person (KYC) or a business (KYB), never both, and the type follows WHO is verified: KYB workflows use the KYB_* features (company paperwork belongs in KYB_DOCUMENTS or DOCUMENT_AI) and must not mix in person-only features (OCR/LIVENESS/FACE_MATCH/NFC/PROOF_OF_ADDRESS/DATABASE_VALIDATION/AGE_ESTIMATION); to verify the people behind a company, create a separate KYC workflow and link it from the KYB_KEY_PEOPLE node. For BRANCHING or conditional logic, use the graph tools instead: didit_workflow_get_field_definitions, then didit_workflow_validate_graph, then didit_workflow_set_graph.
didit_workflow_get3 of 3 parameters describedGet the full configuration of a specific workflow, including `response_attributes` (returned data / data minimization): which data points the client receives per feature.
didit_workflow_update4 of 12 parameters describedUpdate a workflow's top-level SETTINGS only (workflow_label, is_default, retry/expiration, white-label, etc.). Omitted status preserves the current draft/published state; publication changes only when status is explicitly passed. It does NOT change which features run: to add/remove/reorder features, add a conditional branch, or add a Document-AI step, edit the graph with didit_workflow_edit_graph (small ops, preserves the big allow-lists) or didit_workflow_set_graph (full replace) — those are the only ways feature changes actually persist.
didit_workflow_delete3 of 3 parameters describedDelete a draft verification workflow. If the API forbids deletion (including an only or published version), archive it instead. Existing sessions using it are not affected.
didit_workflow_search5 of 5 parameters describedFind verification workflows ACROSS ALL your apps/orgs in one call. Pass `workflow_id` to locate a specific workflow by its version uuid OR stable workflow_id (returns which org/app it lives in), or `search` to match by label. Use this instead of guessing the application when you only have a workflow id. Rows: `workflow_id` = the STABLE id for integrations and sessions; `version_uuid` (= `uuid`) = one exact version, which pins a session to it; `is_live_version` = the version new sessions run. `search` matches the LABEL ONLY: a CAPABILITY (age assurance, AML screening, …) is not searchable by name — for age assurance use didit_workflow_list with `include_age_assurance:true`, which reports what each workflow actually does.
didit_workflow_get_graph4 of 4 parameters describedGet the node/graph for a workflow (the structure: nodes, branches, conditions, Document-AI steps) + `status`/`version`/`is_editable`. Large feature configs (documents_allowed, poa_documents_allowed, phone countries) are SUMMARIZED by default so the response never overflows — set `include_config:true` for the raw config. Includes `returned_data` (response_attributes): the ONLY source of truth for what data the client/relying party receives in the API response and webhooks — never infer that from which features run. Also includes `age_assurance`: whether the workflow checks age, and by which method (document age restrictions and/or AGE_ESTIMATION) — decide that from this block, never from the workflow's name. Pass just `workflow_id`; the owning org/app is resolved automatically. To MODIFY the graph, prefer didit_workflow_edit_graph (small ops, no need to resend big configs).
didit_workflow_get_field_definitions4 of 4 parameters describedList every field you can branch on plus the operators valid for each (e.g. kyc.status, kyc.extra_fields.profession with `fuzzy_match`, document_ai.<key>, aml.risk_score). Call this before building branch rules so the field/operator pairs are valid.
didit_workflow_get_feature_config_schema1 of 1 parameter describedThe exact `config` object each workflow feature accepts - every key, its type, bounds, default, accepted values and meaning. Generated from the backend serializers that validate a save (contract sha256:420b7b24ac86be17505a0966b1b601675ff047dc04eb8ed922035c32f16fc673), so it is never out of date with the API. Call it before configuring a feature you have not configured before, and whenever a config key you set came back missing: a key outside this contract is dropped silently. Pass `feature` for one feature (AGE_ESTIMATION, AML, BANK_VERIFICATION, DATABASE_VALIDATION, DOCUMENT_AI, EMAIL_VERIFICATION, FACE_MATCH, GEOLOCATION, IP_ANALYSIS, KYB_DOCUMENTS, KYB_KEY_PEOPLE, KYB_REGISTRY, LIVENESS, NFC, OCR, PHONE_VERIFICATION, PROOF_OF_ADDRESS, QUESTIONNAIRE, SIGNATURE), or omit it for all of them.
didit_workflow_get_id_verification_methods_catalog4 of 4 parameters describedThe ID Verification methods catalog: per country, whether non-doc lookup (the user types their national ID number and we check the official government database) is offered, which digital-identity wallets (MitID, BankID, itsme, ...) are offered, their availability (`available` can be enabled, `coming_soon` cannot), the plain-language request and response fields, and catalog prices. Read this BEFORE setting `methods` on an OCR node: a method or wallet that is not available for the country is rejected on save. Wallets are an accept-list (no ordering). Every non-document price except the document price is a placeholder until the pricing catalog is published. Pass `country` (ISO3) to get one country's view.
didit_workflow_get_kyb_registry_catalog1 of 1 parameter describedThe KYB registry catalog: per country (ISO-2, unlike `methods`/`documents_allowed` which are ISO-3), which company-data tiers the registries offer - `basic` (Lite: company profile), `shareholders`, `ubo` (beneficial owners) - with their retail price per company selected, and whether continuous registry monitoring is sold there (per company per year). Read this BEFORE writing `kyb_registry_countries_config` with a tier other than `basic`, or switching `kyb_registry_monitoring_enabled` on: a tier the country does not offer is rejected on save, and monitoring only applies to countries at an ownership tier where it is available. A company the applicant types in by hand is billed a flat fee instead. Without `countries` you get a summary (counts, the countries with no registry, the Lite-only ones); pass `countries` for the exact per-country table. Prices here are the catalog's retail prices, not placeholders.
didit_workflow_get_branch_fields5 of 5 parameters describedGiven a candidate graph and a branch node, return the fields actually available at that point (only features that completed on every path reaching the branch, plus dynamically-derived Document-AI/questionnaire fields).
didit_workflow_validate_graph6 of 6 parameters describedDry-run validate a graph WITHOUT saving. Returns per-node errors (bad field/operator, missing dependency, branching on a field before its feature runs, etc.). Always validate before didit_workflow_set_graph.
didit_workflow_edit_graph5 of 5 parameters describedMODIFY an existing workflow's graph with small OPERATIONS - the right tool for editing a live workflow. You send only the deltas; the MCP fetches the full current graph SERVER-SIDE, applies your ops in order, validates, auto-creates a DRAFT (the live version is never touched), and saves. This means huge feature configs (documents_allowed, poa_documents_allowed, phone countries) are preserved exactly and you NEVER resend them. Validation failures return `applied:false` with the errors (nothing saved). The six op shapes are documented on the `operations` parameter; a typical edit upserts new nodes with set_node and rewires their neighbors with set_next. To branch after a feature node, point its `next` at a new branch node whose `branches` carry the rules (from didit_workflow_get_field_definitions) and whose goto targets are new status or feature nodes.
didit_workflow_set_graph5 of 5 parameters describedReplace a workflow's ENTIRE node/graph. Use this only to build a NEW/small workflow from scratch — to modify an EXISTING workflow prefer didit_workflow_edit_graph (you'd otherwise have to resend every node incl. multi-100KB allow-lists). If the workflow is published, a DRAFT is auto-created and the graph applied there (a live version is NEVER mutated); reviewable DRAFT unless `publish:true`. Validate first. Example graph: OCR → branch[ kyc.status==Declined → status(Declined); kyc.extra_fields.profession fuzzy_match 'Software Engineer' score 80 → DOCUMENT_AI(proof of funds) → status(Determine) ; else → status(Determine) ].
didit_workflow_create_draft3 of 3 parameters describedCreate an editable DRAFT version from a published workflow (graph/config edits require a DRAFT). Returns the new draft's uuid.
didit_workflow_publish3 of 3 parameters describedPublish the workflow version that holds the pending changes — makes it live for NEW sessions. Existing sessions are unaffected. Pass the `version_uuid` that set_graph/edit_graph returned: a published version can never hold an edit, so a stable workflow_id resolves to the version that is ALREADY live. Publishing a live version with no draft is refused instead of reported as done.
didit_questionnaire_list4 of 4 parameters describedList all custom questionnaires for your application. One page, newest first: 50 unless `limit` says otherwise, with `count` as the total and `offset` to read the ones after.
didit_questionnaire_create8 of 8 parameters describedCreate a custom questionnaire from an ordered list of questions. Send `title` + `form_elements` (an array of form-element objects, each with an UPPERCASE element_type) — NOT `questions`. The MCP assembles them into the questionnaire graph the backend stores (node ids and their ordering are derived here), so send the questions in the order the user should answer them and never build a graph yourself. Pass status:'draft' to keep it unpublished. A choice list longer than ~100 options does NOT fit in one call (the call is emitted token by token and gets cut off mid-JSON): create with status:'draft' carrying only the first ~100 choices, then add the rest with didit_questionnaire_append_choices in batches (its publish flag publishes with the final batch — a published questionnaire cannot be edited further).
didit_questionnaire_append_choices6 of 6 parameters describedAppend a batch of answer options to ONE question of an existing DRAFT questionnaire (create it with status:'draft' — a published questionnaire cannot be edited). Use it for long choice lists: create with the first ~100 choices, then append the rest here in batches of ~100 until the source list is exhausted, passing publish:true on the FINAL batch to publish. Choices the question already has (same value) are skipped, so retrying a batch never duplicates. Returns a compact summary whose total_choices is the stored count — verify against it instead of re-fetching the questionnaire.
didit_questionnaire_get4 of 4 parameters describedGet questionnaire questions and options. Translations are summarized to one locale by default to keep the response consumable; pass include_translations:true for every locale.
didit_questionnaire_update7 of 9 parameters describedOverwrite a questionnaire's supplied settings. Sending form_elements REPLACES the entire question list, removing omitted questions; pass every question to keep, in order, with UPPERCASE element_type. Omitted status preserves the current draft/published state; publication changes only when status is explicitly passed. Before calling, show the questionnaire and proposed changes and obtain explicit user confirmation, including the full replacement when form_elements is supplied.
didit_questionnaire_delete3 of 3 parameters describedDelete a questionnaire.
didit_vendor_user_list2 of 4 parameters describedList verified users (grouped by vendor_data) for ONE app; with multiple apps (or no scope) it auto-spans all your apps. For a cross-app customer lookup prefer didit_vendor_user_search. Supports limit/offset pagination.
didit_vendor_user_create4 of 11 parameters describedCreate a user record manually (e.g. to pre-register a vendor_data identity or attach allowlisted emails/phones).
didit_vendor_user_get3 of 3 parameters describedGet details of a specific user by their vendor_data identifier.
didit_vendor_user_update5 of 11 parameters describedUpdate a user's profile fields. Only send the fields you want to change. To change ONLY the monitoring status, prefer didit_vendor_user_update_status.
didit_vendor_user_update_status4 of 4 parameters describedSet a user's monitoring status (ACTIVE, FLAGGED, or BLOCKED).
didit_vendor_user_delete5 of 5 parameters describedBatch delete users by vendor_data list, or delete all users. Deletion is permanent. If the API answers 409 (in_filed_report: a selected person is in a finalized regulatory report; subject_deletion_in_progress: another request is still deleting one of them), nothing is deleted: repeat the request once that deletion has finished.
didit_vendor_business_list2 of 4 parameters describedList verified businesses (KYB, grouped by vendor_data) for ONE app; with multiple apps (or no scope) it auto-spans all your apps. For a cross-app lookup prefer didit_vendor_business_search. Supports limit/offset pagination.
didit_vendor_business_create4 of 9 parameters describedCreate a business record manually (KYB).
didit_vendor_business_get3 of 3 parameters describedGet details of a specific business by its vendor_data identifier.
didit_vendor_business_update3 of 9 parameters describedUpdate a business's profile fields. Only send the fields you want to change.
didit_vendor_business_update_status3 of 4 parameters describedSet a business's monitoring status (ACTIVE, FLAGGED, or BLOCKED).
didit_vendor_business_delete4 of 6 parameters describedBatch delete businesses by vendor_data list and/or didit_internal_id list, or delete all businesses. Deletion is permanent. If the API answers 409 (in_filed_report or subject_deletion_in_progress), nothing is deleted: repeat the request once the other deletion has finished.
didit_transaction_list2 of 4 parameters describedList monitored transactions for ONE app; with multiple apps (or no scope) it auto-spans all your apps. For a cross-app query prefer didit_transaction_search. Supports limit/offset pagination.
didit_transaction_create9 of 13 parameters describedSubmit a transaction for monitoring and rule evaluation. transaction_details and subject shapes depend on transaction_category.
didit_transaction_get3 of 3 parameters describedGet a single monitored transaction and its rule-evaluation result.
didit_transaction_screen_wallet5 of 5 parameters describedScreen a single crypto wallet address for AML risk WITHOUT creating a transaction. Returns risk_score, severity, sanctions_hit, source/destination of funds. Requires transaction monitoring to be configured (a provider key) or returns 409.
didit_transaction_rule_list3 of 10 parameters describedList transaction-monitoring rules for one app. Filter by source (PRESET or CUSTOM), category, mode (ACTIVE evaluates and acts, TEST evaluates and records but skips actions, DISABLED is skipped), bundle, or free-text search. ordering accepts one or more comma-separated fields; prefix a field with '-' to reverse it.
didit_transaction_rule_get3 of 3 parameters describedGet one transaction-monitoring rule, including its persisted conditions, aggregation, scope, actions, and counters.
didit_transaction_rule_create11 of 12 parameters describedCreate a CUSTOM transaction-monitoring rule. Include the entire requested persisted rule in this call: conditions, aggregation checks, and actions. Backtest accepts a hypothetical config but never attaches aggregation or actions to a created rule. Prefer mode:'TEST', backtest the exact saved config, then switch to ACTIVE only after confirming the hit rate. title is unique among non-deleted rules in the app. actions is REQUIRED: every outcome the user named (review, decline, approve, score, case) must appear there; pass [] ONLY when the user explicitly wants a monitor-only rule that does nothing on match. To change a library (PRESET) rule, send its definition as a copy with metadata.based_on (the library key) and metadata.based_on_rule_uuid.
didit_transaction_rule_update8 of 13 parameters describedPartially update a transaction-monitoring rule. Send every changed persisted field in this PATCH. PRESET rules only allow mode changes: to change a PRESET rule's conditions, create a CUSTOM copy with didit_transaction_rule_create (the library rule's definition plus metadata.based_on = its library key and metadata.based_on_rule_uuid), then set the library rule's mode to DISABLED so a transaction does not alert twice. Before raising impact, backtest the exact candidate config and keep the rule in TEST until the result is acceptable.
didit_transaction_rule_delete3 of 3 parameters describedDelete a CUSTOM transaction-monitoring rule. PRESET rules must be removed with didit_transaction_rule_uninstall.
didit_transaction_rule_backtest7 of 7 parameters describedDry-run a hypothetical rule config against up to the 10,000 most recent transactions in the app. This writes nothing, accepts no rule_uuid, and never changes or completes a saved rule. To test a saved rule, read it first and pass its exact conditions, aggregation, evaluation_mode, and scope here.
didit_transaction_rule_library_list2 of 7 parameters describedList Didit's preset rule library. Each entry includes its full config, bundle, tags, industries, and is_installed. Filter by bundle, category, or search; paginated.
didit_transaction_rule_install4 of 4 parameters describedInstall PRESET rules by library_keys, bundle, or both. If both are supplied, only keys in that bundle are installed. At least one selector is required. Installed rules start ACTIVE when transaction monitoring is enabled for the app, otherwise DISABLED. Set TEST explicitly before relying on a newly installed rule.
didit_transaction_rule_uninstall4 of 4 parameters describedRemove installed PRESET rules by library_keys, bundle, or both. If both are supplied, only keys in that bundle are removed. At least one selector is required. To pause without removing, update mode to DISABLED.
didit_org_get_balance1 of 1 parameter describedGet current credit balance and auto-refill settings. Every check is billed from the first one: from 1 Nov 2026 an eligible Free organization gets $10 of monthly credit usable on any feature, paid plans carry their monthly plan credit (Growth $115, Scale $350), and no workflow is free of charge. The Free credit is decided per organization: `free_credit.receives` says whether this one gets it, and `free_credit.this_month` holds this month's credit, also once used up. `free_credit.eligibility` says why an organization does NOT get it, because the credit needs a work email: personal_email means it signed up with a personal email, temporary_email with a temporary one, company_exists that its company already has an organization, api_created that it was created through the API. Such an organization never had the credit, so it was not used up. Answer from these fields, never from the plan alone; when `free_credit` is absent, say you could not read it. Never answer a "not enough credits" report from this result alone: read the failing workflow with didit_workflow_list first - `is_white_label_enabled: true` adds $0.20/session, which is a common cause of credit running out faster than expected. This result also carries `allow_free_usage` (organizations on negotiated terms keep the monthly free allowance they were sold until renewal; false means that allowance is off for this organization) and `usage_summary.white_label_sessions`. When the balance cannot cover what the person wants to run, give them `next_steps.top_up_url` (the console top-up dialog with the amount filled in; replace the number to change it) or `next_steps.automatic_top_up_url` (save a card and turn on automatic top-up so the balance never runs out). Didit tools never charge a card. Meanwhile keep building in the organization's sandbox application (`mode: sandbox` in didit_context_get), which is never billed.
didit_org_top_up5 of 5 parameters describedTop up credits ($50 to $20,000 per top-up). payment_method "checkout" (default) creates a Stripe Checkout page and returns checkout_session_url for the person to pay; it never charges a card by itself. payment_method "saved_card" charges the organization's saved primary card off-session, now, when a person turned on "Connected AI agents can top up with your saved card" in the Didit console (Billing > Payment methods; off by default; didit_org_get_balance shows it as agent_top_up.enabled where this tool is offered). It needs idempotency_key: choose ONE unique key per intended top-up (for example a UUID) and reuse that same key on any retry of that same top-up, so a retry can never charge twice; never reuse a key for another amount and never retry with a new key. A refusal carries `code`, `detail` and `next_step` {action, url, description}: agent_top_up_not_enabled (the switch is off), authentication_required (the card's bank asks the cardholder to approve), payment_method_required (no saved card) and card_declined are finished by the person: give them next_step.url (a Top up dialog link that says which agent asked, with the amount filled in).
didit_billing_get_plan1 of 1 parameter describedGet the organization's plan: the plan and its status, monthly fee and credit, the next charge, the last payment, any unpaid invoice and any scheduled plan change. `next_steps.plans_url` is the console page where the person changes the plan.
didit_billing_quote_plan3 of 3 parameters describedQuote a plan change without making it: what it charges today (`amount_due`), the credit it adds (`credit_added`) and the billing period. Plans: free, growth ($99/month, $115 of credit a month, 4% off every price), scale ($299/month, $350 of credit a month, 10% off); yearly billing is cheaper. Show the person the quote; they confirm and pay on `next_steps.plans_url` (Didit tools never buy a plan).
didit_billing_list_payment_methods1 of 1 parameter describedList the organization's saved cards (brand, last 4 digits, expiry) in the order automatic charges try them, card 1 being the primary. An empty list means no card is saved: top-ups then go through Checkout or the console. `next_steps.add_card_url` is where the person adds a card or turns on automatic top-up.
didit_billing_list_receipts3 of 3 parameters describedList the organization's payment receipts (top-ups and plan charges), newest first.
didit_billing_get_details1 of 1 parameter describedGet the billing details printed on every receipt and invoice: legal name, tax ID, address, billing emails and PO number.
didit_billing_update_details7 of 7 parameters describedChange the billing details printed on receipts and invoices. Pass only the fields to change; the rest keep their saved values. Before calling, show the person each change and get their confirmation. When the organization confirms billing changes with an authenticator code, the change is refused and the person saves it in the console instead (the error names the page).
didit_branding_get2 of 2 parameters describedGet the current branding customization (logos, colors) applied to your verification UI.
didit_branding_update9 of 9 parameters describedUpdate verification UI branding images and/or the SMS sender name. Each image is a local absolute *_path (local/stdio runs) or inline *_base64 content (hosted runs).
didit_webhook_list2 of 2 parameters describedList configured webhook destinations. Each destination has its own URL, version, enabled flag, subscribed events, and redacted signing-secret metadata.
didit_webhook_create7 of 7 parameters describedCreate a webhook destination. The signing secret is redacted in the response.
didit_webhook_get3 of 3 parameters describedGet a single webhook destination with redacted signing-secret metadata.
didit_webhook_update3 of 8 parameters describedOverwrite the supplied settings of an existing webhook destination. Changing its URL redirects future deliveries; disabling it stops deliveries, and subscribed_events replaces its event subscriptions. Before calling, show the destination UUID and each proposed field change and obtain explicit user confirmation.
didit_webhook_delete3 of 3 parameters describedDelete a webhook destination so it stops receiving events.
didit_lists_list4 of 6 parameters describedList all lists (blocklists, allowlists, custom) for the application. Filter by list_type or entry_type.
didit_lists_create5 of 6 parameters describedCreate an allowlist or custom list. System blocklists are auto-provisioned (one per entry type) and cannot be created. name must be unique per application.
didit_lists_get3 of 3 parameters describedGet a single list's details by UUID.
didit_lists_update3 of 5 parameters describedRename or update the description of an allowlist/custom list. System blocklists are immutable.
didit_lists_delete3 of 3 parameters describedDelete an allowlist/custom list and all its entries. System blocklists cannot be deleted.
didit_lists_entries_list5 of 7 parameters describedList entries in a specific list. Use search to filter by value or label.
didit_lists_entry_create11 of 11 parameters describedAdd an entry to a blocklist/allowlist/custom list. Pass value directly, or reference_session_id to auto-extract from a session (face, document, phone, email, IP, device). Pass both to disambiguate when a session has multiple values of the same type. For the face blocklist, face_image_source picks which of the session's faces to block: the liveness selfie (default) or a selfie the front camera took while the user photographed their document. For face entries without a session, use didit_lists_entry_upload_face instead.
didit_lists_entry_upload_face5 of 7 parameters describedAdd a face entry to a face list by uploading an image directly (no session needed). Use this when you have a photo but no reference_session_id. Provide the image as image_path (local/stdio runs) or image_base64 (hosted runs).
didit_lists_entry_delete4 of 4 parameters describedRemove an entry from a list. Also unblocks the underlying user/business if applicable.
didit_verify_id10 of 10 parameters describedVerify an identity document by submitting front (and optionally back) images. Returns structured OCR data and authenticity checks. Each image is a local absolute *_path (local/stdio runs) or inline *_base64 content (hosted runs).
didit_verify_poa7 of 9 parameters describedProof of Address verification. Submit a single document image to extract and validate address information. The image is a local absolute *_path (local/stdio runs) or inline *_base64 content (hosted runs).
didit_verify_account_check7 of 7 parameters describedCheck that a bank account belongs to the person or company you name, before sending money to it. No session: send the name and the account, read the result. result is match, close_match, no_match or cannot_verify; on a close_match, holder_name is the name the bank holds the account under (when returned); account comes back masked to its last 4 characters. Billed once per check the bank answers at the account's country price, whatever the result (no_match and cannot_verify too); an error is never billed. Every call is a NEW billed check (no idempotency key, no read-back): do not retry blindly and store the response. Errors: 400 names the field and reason (consent required: in countries whose check needs the account holder's explicit consent, ask for it and send consent:true; nothing is billed); 402 top up; 404 country_not_covered or bank_not_covered; 503 the bank could not answer, send again later. Where Bank Verification is not switched on (production at the moment) it answers 404 bank_verification_not_available. A sandbox application calls no bank and charges nothing: the last 4 characters of the IBAN, else the account number, choose the answer (0001 close_match with holder_name, 0002 close_match without, 0003 and 0006 no_match, 0004 cannot_verify, 0005 a 503, 0009 a 400, anything else match).
didit_verify_database5 of 19 parameters describedValidate identity data against national and global authoritative data sources. issuing_state (ISO 3166-1 alpha-3) is required and selects which sources run.
didit_verify_kyb_search7 of 9 parameters describedSearch official company registries for a business by name and/or registration number. Returns candidate matches; pass a candidate's kyb_response_id to didit_verify_kyb_select to pull the full record.
didit_verify_kyb_select3 of 6 parameters describedResolve a candidate from didit_verify_kyb_search into a full company registry record. kyb_response_id is the per-search candidate handle returned by the search. A resolved record carries registry_status and verification_status (verified, failed or unknown).
didit_verify_passive_liveness5 of 5 parameters describedPassive liveness detection -- verify a person is physically present from a single image (no interaction required). The image is a local absolute image_path (local/stdio runs) or inline image_base64 content (hosted runs).
didit_verify_face_match7 of 7 parameters describedCompare two facial images to determine if they belong to the same person (1:1 face matching). The score is symmetric, so image order does not affect the result. Each image is a local absolute *_path (local/stdio runs) or inline *_base64 content (hosted runs).
didit_verify_face_search5 of 5 parameters describedSearch for a face against a database of previously verified faces (1:N face matching). The image is a local absolute image_path (local/stdio runs) or inline image_base64 content (hosted runs).
didit_verify_age5 of 5 parameters describedEstimate a person's age from a facial image. Also performs passive liveness check. The image is a local absolute image_path (local/stdio runs) or inline image_base64 content (hosted runs).
didit_verify_aml8 of 11 parameters describedAML screening against global watchlists, PEP lists, sanctions, and (optionally) adverse media. Supports person and company entity types.
didit_verify_email_send3 of 3 parameters describedSend a one-time verification code to an email address. Code valid for 5 minutes.
didit_verify_email_check3 of 4 parameters describedVerify the OTP code sent to an email. Max 3 attempts per code.
didit_verify_phone_send3 of 3 parameters describedSend a one-time verification code to a phone number. Delivered on WhatsApp by default, falling back to SMS where WhatsApp cannot reach that country or number. The code is valid for 5 minutes. Returns request_id plus status: Success (a new code was sent), Retry (resent while an earlier code for that number is still pending) or Blocked (the number was refused; see reason). Paid from the organization's balance, so a live application really delivers a message, and answers 403 until the organization's first top-up or when the balance is short; a sandbox application mocks the send for free. Repeated sends to one number can answer 429 with code PHONE_RATE_LIMITED. Sends with the application's API key, so the caller's role needs create:sessions (Owner, Admin or Developer; an organization owner or admin with a local didit_account_login token), and an MCP sign-in needs the didit:management scope to read that key.
didit_verify_phone_check3 of 4 parameters describedCheck the one-time code sent to a phone number and return what Didit knows about that number. Maximum 3 attempts per code, and the code expires 5 minutes after it was sent. Returns status Approved, Failed (with the attempts remaining), Declined, or 'Expired or Not Found' when that number has no code pending. A sandbox application mocks the check and approves any code. Sends with the application's API key, so the caller's role needs create:sessions (Owner, Admin or Developer; an organization owner or admin with a local didit_account_login token), and an MCP sign-in needs the didit:management scope to read that key.
didit_blocklist_get5 of 5 parameters describedList the application's blocklist, newest first: the faces, identity documents, phone numbers and email addresses that are blocked, each with the `session_id` it was blocked from (what didit_blocklist_remove takes). One page: 50 entries unless `limit` says otherwise, with `count` as the total and `offset` to read the ones after.
didit_blocklist_add7 of 7 parameters describedBlock what a verification session holds, so that later sessions of this application with the same item are caught by the blocklist check: its face, its identity document, its phone number or its email address. Name the session and set at least one of the four flags to true. A flag works only when the session has that item. Undo with didit_blocklist_remove.
didit_blocklist_remove7 of 7 parameters describedLift a block. Name the session the item was blocked from (the `session_id` of the entry in didit_blocklist_get) and set at least one of the four flags to true.
didit_allowlist_add4 of 4 parameters describedMark the phone number of a verification session as trusted in this application, so that phone verification does not turn it away as high risk. A phone number is the only thing that can be allowlisted, and a blocked number stays blocked. Name the session and set allowlist_phone to true.
didit_case_list2 of 4 parameters describedList investigation/compliance cases for ONE app; with multiple apps (or no scope) it auto-spans all your apps. For a cross-app query prefer didit_case_search.
didit_case_get2 of 3 parameters describedGet a case with its details.
didit_case_create12 of 12 parameters describedOpen a case for a compliance review. A case is about a subject (one of your users or businesses) or is linked to at least one session or transaction: pass `subject`, `links` or both. `subject.uuid` is the `didit_internal_id` of the user or the business, as didit_vendor_user_get and didit_vendor_business_get return it; it is not their vendor_data. Returns the case with its `uuid`. With review_duplicates true the API first looks for cases already open about the same subject and answers 409 case_duplicate_review_required when it finds one: show it to the person, and to open a separate case anyway send separate_incident_reason.
didit_case_statistics8 of 10 parameters describedCase analytics for a period (totals, series over time, the queue, outcomes and workload): the last 30 days unless both dates are given, two years at most. Every filter narrows the numbers.
didit_case_manage4 of 6 parameters describedAct on a case: assign, resolve, reopen, escalate, comment, or update.
didit_investigator_queue_list8 of 11 parameters describedList the Investigator's open reports (the review queue), Ready to accept first. Each row has run_id, the item, the recommendation (approve/decline/escalate/request), confidence and ready_to_accept. Read the full report with didit_investigator_report_get.
didit_investigator_report_get3 of 3 parameters describedGet one Investigator report: recommendation, confidence, risk, narrative, the basis (policy rules), evidence with its sources, guardrails, precedents (similar past decisions), questions and provenance (model, prompt hash, policy version). Use it to explain why the Investigator recommended something.
didit_investigator_accept4 of 5 parameters describedAccept the Investigator's recommendation on one item: the item is decided through its normal route (approve, decline, request information or escalate) and logged in the Investigator Activity under the user's name. A changed narrative makes it 'edited and accepted'. Requires human approval: describe the exact action to the user, wait for their explicit confirmation, then call with confirm:true.
didit_investigator_override5 of 7 parameters describedOverride the Investigator: decide one item with another action, a category and a reason (10-500 characters). Also overrides a decision the Investigator automated while it still stands; repeated overrides become policy questions. Requires human approval: describe the exact action to the user, wait for their explicit confirmation, then call with confirm:true.
didit_investigator_batch_accept4 of 4 parameters describedAccept up to 50 ready recommendations at once; each item is its own decision and one that cannot be accepted is skipped with its reason. List the items and their recommendations to the user first. Requires human approval: describe the exact action to the user, wait for their explicit confirmation, then call with confirm:true.
didit_investigator_add_note4 of 4 parameters describedRecord a fact the user adds about a report's item as a team note (any kind: verification, business, transaction, case). The note shows with the report and the Investigator reads it on its NEXT review of the item; it does not re-evaluate on its own. To have it review the item again now, use didit_investigator_rereview (it can carry the note itself).
didit_investigator_rereview5 of 5 parameters describedAsk the Investigator to review an item again (for example after the user added a fact): a fresh review starts in the background and replaces the item's open report when it starts. The usual stage, spending limit and guardrails apply and the Investigator never decides that review on its own. Only the item's open, finished report can be re-reviewed (409 not_rereviewable), and one at a time (409 rereview_pending). Read the new report with didit_investigator_queue_list / didit_investigator_report_get once it is ready. Requires human approval: describe the exact action to the user, wait for their explicit confirmation, then call with confirm:true.
didit_investigator_record_get4 of 4 parameters describedGet the Investigator's review of one verification (session_id, KYC or KYB) or one transaction (transaction_id): its newest run, open or decided, with run_id, status, recommendation, confidence, summary, whether it was automated and who decided it; null when it never had one. Use it for a record the queue does not list (decided items, a record page). Read the full report with didit_investigator_report_get once status is completed. A case's run is in didit_case_get's investigator.
didit_investigator_question_list4 of 5 parameters describedList the Investigator's policy questions (situations its policy does not cover, or rules the team keeps overriding), newest first, with counts. One page of 50 unless `limit` says otherwise: `count` is the total and `offset` reads the ones after.
didit_investigator_question_get2 of 3 parameters describedGet one policy question with the team's last 30 days on that situation and the four possible answers.
didit_investigator_question_answer4 of 6 parameters describedAnswer a policy question: approve, decline, request (ask for more information) or person (always ask a person). Any answer but 'no change' writes a new policy version and re-checks the waiting reports of that queue. Requires human approval: describe the exact action to the user, wait for their explicit confirmation, then call with confirm:true.
didit_investigator_question_dismiss3 of 4 parameters describedClose a policy question without an answer; the policy stays as it is. Requires human approval: describe the exact action to the user, wait for their explicit confirmation, then call with confirm:true.
didit_investigator_backtest_start5 of 5 parameters describedStart a FREE deterministic backtest: replay the current policy (or the setup or SOP draft) on the last 30 or 90 days of decisions and get agreement per queue and tier, with sample warnings. Decides nothing. Read it again with didit_investigator_backtest_get.
didit_investigator_backtest_get2 of 3 parameters describedGet a stored free backtest: agreement for every queue, tier and threshold, and its sample warnings.
didit_investigator_performance_get3 of 4 parameters describedGet the Investigator's performance: agreement with the team, volume, time saved, calibration, per-queue numbers and overrides over 7, 30 or 90 days.
didit_investigator_dashboard_get3 of 3 parameters describedGet the Investigator dashboard: 30-day automation, weekly history, spend, overrides and QA since go-live.
didit_investigator_activity_list8 of 9 parameters describedList the Investigator's append-only Activity log, newest first: what the agent and people did (reports, decisions, overrides, autonomy and guardrail changes), each with its actor and provenance.
didit_report_list7 of 7 parameters describedList the application's generated reports, newest first, with how many there are of each status (`counts`). One page: 50 unless `limit` says otherwise, with `count` as the total and `offset` to read the ones after.
didit_report_get2 of 3 parameters describedGet a report's status/details.
didit_report_get_download_url2 of 3 parameters describedGet a signed download URL for a finished report.
didit_report_export3 of 4 parameters describedStart an export report. kind ∈ sessions | transactions | businesses | vendor-users | vendor-businesses. `data.columns` is a kind-specific schema — for sessions/vendor-users/vendor-businesses an unknown column 400s with `Invalid columns`, and that error's `allowed` field carries the full valid-column list for that kind, so retry from the error alone rather than guessing names like `date`/`status`/`total_price`.
didit_audit_log_list11 of 11 parameters describedList the organization's audit log, newest first: who did what in the organization (members, API keys, Didit itself, the Didit assistant and the Investigator). Each entry has actor {type, name, email}, action, description, target {type, id, label} and changes. Decisions the Investigator took on its own are entries with actor_type 'investigator' and changes.investigator {run_id, outcome, recommendation, confidence, policy}. A person's decision on an Investigator report is that member's entry with changes.investigator.outcome 'accepted' or 'overridden'. To answer "who approved session 20458", pass search='20458' (or target_id=<the record's uuid>) and read the actor of the decision entry.
didit_alert_list2 of 2 parameters describedList the application's alerts, one per channel: `alert_type` (email or slack), where it is sent (`value`) and whether it is on (`is_active`).
didit_alert_configure4 of 4 parameters describedSet where one of the application's alerts is sent, or turn it on or off. `alert_type` is the channel. An alert with no destination is always off. If the API answers 409 legacy_alerts_read_only, this application's alerts are managed in the console's Notification settings and cannot be changed here.
didit_org_list_members1 of 1 parameter describedList the organization's members.
didit_org_invite_member4 of 4 parameters describedInvite up to 5 members to the organization (emails + role + the applications they may access).
didit_org_update_member4 of 4 parameters describedUpdate a member's role and the applications they may access.
didit_org_remove_member2 of 2 parameters describedRemove a member from the organization. It cannot be undone: to come back the person has to be invited again.
didit_org_list_roles1 of 1 parameter describedList the roles available in the organization.
didit_org_create_api_key7 of 7 parameters describedCreate the API key the integration you are building needs (sent as x-api-key to the verification API), for one application. Didit allows ONE live key per connected app and application, scoped to what the signed-in person can do (never full access, never API key management), rate-limited and audit-logged; the person sees it with this connected app in the console and can revoke it there. The secret is returned ONCE: write it straight to the project's secret store or .env and never repeat it in the chat. Pick the application by its mode in didit_context_get: a sandbox key to build and test for free, a live key to go live. Requires confirm:true after the person agreed.
didit_org_list_api_keys2 of 2 parameters describedList one application's API key metadata with raw key values redacted. API keys belong to an application, so application_id is required: read it from didit_context_get.
didit_workflow_build_graph9 of 9 parameters describedBuild a complete workflow graph from a PLAIN FEATURE SPEC in one deterministic call - regulations are never consulted. Gate-then-commit: the result is either an accepted build (graph_summary + spec: materialize with ui_workflow_apply_graph {spec} on an open editor, or include_graph:true + didit_workflow_set_graph headless), or 'unsupported'/'questions' naming exactly what cannot be built or what to ask the user FIRST - in that case build NOTHING and relay them. Never rebuild the returned graph by hand and never retry with guesses. Read-only: persists nothing. document_rules express per-country/document filtering, including US state-scoped subtypes (e.g. exclude the Nevada driver's license) and the country wildcard 'ALL' (every catalog country: 'only passports, from any country' = one rule {country:'ALL', document:'P', action:'only'}).
didit_docs_search3 of 3 parameters describedSearch the Didit documentation (docs.didit.me): API reference, webhooks, SDKs, verification features, pricing and guides. Use it for any question about how Didit works or how to integrate it, instead of answering from memory. Returns the matching sections with their titles and links; cite the link you used.
Tested in a sandbox with an empty environment: no API keys, a throwaway folder, nothing from a real account. A passing test shows the server runs and responds. It is not a security audit.