All servers / Developer / Telora MCP server
Telora MCP server
MCP server exposing Telora product operations to Claude Code
Tested by ConnectATDScore 82 out of 100. Tested 2026-11-19 on package version 0.23.30. Package: @telora/mcp-products · source
Score breakdown
| Runs | 18 / 20 | |
|---|---|---|
| Tool descriptions | 20 / 20 | |
| Package safety | 12 / 20 | |
| Maintenance | 20 / 20 | |
| Adoption | 12 / 20 |
What we saw
- Starts and answers the MCP handshake: yes (6.1 seconds including download)
- Tools offered: 15
- Last published: 2026-11-15
- Weekly downloads (npm): 1,482
- License: MIT · Direct dependencies: 4
Notes
- The package runs install scripts, which execute code when you install it.
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.
telora_product23 of 23 parameters describedProduct registry: manage the portfolio of software products. Each product scopes focuses, deliveries, issues, and context documents. Actions: list products, create a new product (name required), get product details (by name or id), update product settings (vision, audience, QA services, QA target), overview (full product summary with focuses and deliveries), QA TARGET: qaTargetEnvironment / qaTargetBaseUrl / qaTargetAccessNote declare where a human goes to test the product's built UI -- one per product. QA packets (telora_product_focus qa_get) write case paths RELATIVE to qaTargetBaseUrl (http/https), so a human can open each case directly; null clears a field. get returns all three. register_local_endpoint (DEPRECATED -- use `telora_local_model create` instead. Retained for back-compat only: it still writes the legacy single per-product local_model_coordinates blob, which the org-scoped `local_models` registry supersedes. Registers non-secret local-model coordinates {baseUrl, model, wireApi} + a per-product engine binding; the API key stays local and is NEVER sent or stored). List responses include a 1-based `index` on each item (response-local; recomputed per call).
telora_product_delivery31 of 31 parameters describedWork packages: individual shippable increments within a focus. Each delivery has acceptance criteria, technical context, and tracks progress through issues. Actions: list deliveries (filter by status/focus), create a delivery (name + productId; injectionNodeId anchors the delivery to a reality-tree injection -- REQUIRED when the focus is past planning and has an active reality tree: the create is REFUSED (error, no row) if omitted, and likewise refused on a past-planning focus that has no tree. Anchoring is the default path; for a genuinely tree-less delivery you must pass BOTH allowUnanchored=true AND unanchoredReason (a bare override with no reason is refused). The converse also holds: an injection already materialized by a live delivery REFUSES a second one (duplicate_injection_anchor, no row) -- put residual work on the EXISTING delivery it names, or for a genuinely deliberate split pass BOTH allowDuplicateAnchor=true AND duplicateAnchorReason. each of these two is an effortful, reason-bearing, attributed override, and a bare boolean with no reason is refused (override_reason_required). AIM AT CAUSES: the injection a delivery anchors to targets the CAUSES the delivery changes, never an undesired effect. Write the delivery's acceptance criteria as those causes' declared futures -- what verify judges it on; the undesired effects the causes EARN are the focus's criteria, judged at the focus review once every delivery that earns them has landed, so no delivery is held to a sibling's work. REVIEW-FINDING RULE: no delivery is created in answer to a BLOCKING review finding. Only an OPEN review-filed finding the review classified blocking (a not_met criterion names it) refuses: while one is open, an agent create on that focus, or an update attaching a delivery to it, is REFUSED (review_finding_open, no row), naming those findings -- close a shortfall as a Bug or Context Group on the delivery it was filed against. An open CONCERN (no not_met criterion names it) refuses nothing: a create that names no finding is admitted and every concern stays where it was filed. Taking a concern is OPTIONAL: a create MAY name concerns in takesFindingIssueIds when it anchors an injection no live delivery holds (the new gap's own cause/effect/injection), and they move to the new delivery -- a blocking finding never moves, and taking a concern never gets a create past an open blocking finding. A focus a concern's Radar blip (or its theme) was PROMOTED to may take that concern even though ANOTHER focus's review filed it -- while the filing focus still records it as a concern; it then moves here and is this delivery's owed work. New work found after a delivery is done goes HERE, never as a Task on the done delivery (refused). RECORD/QUEUE RULE (an armed focus): the create is never refused for being out of scope. On a focus a human has ARMED, a delivery whose injection serves no effect the arm covers is CREATED and RECORDED -- it rests at execution_status 'planning', is invisible to the running team, and executes when the focus is next armed. The response says so and names the act that would make it run NOW instead: draw a causal edge (sufficient or and) from one of its injection's targets to an effect the arm already covers, which makes it means rather than new intent. Nothing is refused and nothing is frozen; do not re-try the create with different wording. An UN-ARMED focus (no assigned agent role) runs no team until a human arms it, so a create there that would land queued (the past-planning default, or an explicit queued/coding/verify) is RECORDED the same way: it rests at 'planning', the response says why (recorded_for_next_arm), and arming the focus promotes it to queued; an update moving a planning delivery to a runnable status on an un-armed focus likewise leaves it at planning, while scoping statuses (planning, draft, cancelled) apply as asked; declare workFacing 'outward'|'inward' at create to record whether the delivery opens new ground or r…
telora_product_focus40 of 40 parameters describedExecution roadmap: group deliveries into themed work scopes. Focuses define what the AI daemon works on -- assigning an agent role queues the focus for autonomous execution. Actions: list focuses for a product, create a focus (name + productId), update focus settings (description, status, agent role, pipeline config), reorder focus priority (focusIds in rank order; the response is the reordered rows ONLY -- the ids given, in their new rank, at `detail` (default summary), with notFoundIds naming any id that is no focus of the product -- for the surrounding order, call list), merges (list merge history for a focus), retrospective_trigger (create a pending retrospective for a focus), retrospective_run (execute a pending retrospective via LLM), retrospective_list (list retrospectives for a focus), retrospective_get (get a single retrospective by ID), review_complete (terminal action for a review agent: writes a focus_reviews row with outcome + summary + references, and updates focus.last_reviewed_at and focus.last_review_outcome. A finding blocks only when a not_met criterion names it (acVerdicts[].issueIds); everything else the review files is recorded as a concern; a not_met reopening a criterion an earlier review under the same arm judged met with evidence blocks only when its acVerdicts[].falsifier test fails at the judged commit; a review whose verdicts cover every criterion with none not_met is recorded as approved; each concern is captured as a maintenance blip on the Radar in the same write, and concernPlacements places it among the themes there), review_list (read completed focus reviews back, newest first, with the decisions taken under uncertainty each one recorded -- question, options, chosen, why reversible, what changes if the human decides otherwise; scope by focusId, productId, or omit both for the whole organization; carries the decision-record rate: the share of completed reviews that recorded at least one resolved fork), review_get (read one focus review by focusReviewId: full summary plus its decisions rendered for a human), criteria_standing (the focus's FLOOR, read without running a review: what this focus is FOR, from BOTH statements of it, and where each part stands. Every entry carries source: 'tree' criteria are the futures (frt_statement) of the undesired effects the focus's ARM covers, each with its nodeId, seq, drafted (a MACHINE wrote that future and a human accepted it by arming) and landing; 'declared' criteria are the lines of the focus's own acceptance_criteria field. THE TWO ARE JUDGED BY DIFFERENT RULES, and status says which fired: a tree criterion reads 'met' ONLY when VERIFY RECORDED its node's future as landed -- reality, read from the node row, never from any text and never from a delivery targeting the node -- while a declared criterion reads 'addressed' when a current delivery DECLARES it. Met and addressed are different claims (reality vs scoping); 'open' is everything else, and every open criterion is echoed verbatim. treeBasis says WHOSE statement of the bar you are reading: armed_stamp (the arm's own armed_baseline_criteria, verbatim -- THE bar, immune to a later frt_statement edit), derived_live_unstamped_arm (armed before the stamp existed, so derived live and APPROXIMATE), derived_live_unarmed, or none. On an UN-ARMED focus the tree half is a PREVIEW of what an arm would stamp and what it would REFUSE on: floorBasis reports derived_live_unarmed -- never a floor, because nothing was stamped to reach -- and the read adds unstatedCoveredEffects (covered effects with NO stated future, which is what the arm refuses on, named by the same seq the refusal message names) and draftedCoveredEffects (covered effects whose future is only a machine draft: stamped, not refused). Consult it before adding another delivery to a focus that may already be close to met, and before arming one. Available BETWEEN reviews and on a focus that has NEVER been reviewed: it is a read, not a review act, so it ne…
telora_product_strategy42 of 42 parameters describedDEPRECATED: use telora_product_focus. This alias accepts both legacy strategyId/strategyIds and new focusId/focusIds parameters and will be removed in a future release.
telora_product_focus_debug5 of 5 parameters describedWRITES agent-authored, human-facing visualizations on request (the focus Debug tab) -- it is NOT the canonical store of current architectural reality. To READ current reality during scoping, use telora_reality_tree get/dump (and telora_reality_tree_node/_edge list), not this tool. Focus debug artifacts: agent-written code-reality views (Mermaid graph + markdown panels) rendered on the focus page Debug tab. One row per (focusId, sectionKind). Section kinds: mermaid, file_inventory, git_stats, test_status, lint_typecheck, ac_verification. (Two sections are DAEMON-written and absent from this enum -- do not author or clear either: integration_sync, the focus branch conflicts with integration; and directive_undelivered, a stage directive its target session ended mid-delivery never received.) Actions: get(focusId) returns all sections + updatedAt; set(focusId, sectionKind, content) upserts one section; clear(focusId, sectionKind?) deletes one section (sectionKind given) or all (omitted).
telora_product_issue19 of 19 parameters describedTask tracking: manage issues (tasks, bugs, context groups) within deliveries. Issues flow through To Do -> In Progress -> Done. Context groups organize related tasks with shared files and AI context. Actions: list issues (by delivery or product-wide), create an issue (title + deliveryId), update issue fields (status, priority, description), delete an issue. List responses include a 1-based `index` on each item (response-local; recomputed per call).
telora_infra_environment10 of 10 parameters describedInfrastructure environments: named deployment targets (production, staging, dev) within a product. Each environment groups infrastructure assets and has a status (active, provisioning, decommissioned). usersUrl is the address this environment's USERS open (absolute https): release notes build their links on it, and a human cuts each release (telora_release) for one environment. When usersUrl is unset, the product's QA target base URL counts only for the environment the QA target names. Actions: list (by product), create (name required, unique per product), update, delete. List responses include a 1-based `index` on each item (response-local; recomputed per call).
telora_infra_asset14 of 14 parameters describedInfrastructure assets: typed resources (database, container, service, dashboard, storage, proxy, dns, monitoring, ci_cd) within a product. Assets can optionally belong to an environment and store connection details (url, ip, port). Actions: list (by product, optional environment/type filter), create, update, delete. List responses include a 1-based `index` on each item (response-local; recomputed per call).
telora_infra_relationship10 of 10 parameters describedAsset relationships: directed links between infrastructure assets within the same product. Types: connects_to, deploys_to, proxies, depends_on, hosts. Actions: list (by product), create (source + target + type), delete. List responses include a 1-based `index` on each item (response-local; recomputed per call).
telora_infra_secret5 of 5 parameters describedInfrastructure secrets: encrypted key-value pairs stored per asset using Vault. List and create show names only (no values). Use get to decrypt and reveal a secret value on demand. Actions: list (by asset), create (name + value), get (decrypts), delete. List responses include a 1-based `index` on each item (response-local; recomputed per call).
telora_context_assemble6 of 6 parameters describedAssemble layered context for a product, focus, or delivery. Returns pre-formatted markdown with product overview, focus/delivery scaffolding, a wiki section drawn from wiki_search using scope-derived queries (or your own `queries` override), and an 'Actions on this <scope>' section enumerating the primary MCP tool + HTTP fallback for each capability available on the entity. Pass an `assembly` recipe to select/compose specific sections over the DB-backed resolver subset, or `modelContextTokens` to lift the wiki budget to your model's context window. Use at session start for architectural/design work.
telora_connector_start3 of 3 parameters describedConnect this directory to Telora and run the daemon for it. In a directory not yet bound to a product (even when a productId is passed), from a personal-tracker session, or with no Telora credentials yet, it runs the browser-approved connect: the first call opens a connect request and returns a URL + short code for the user to approve in the web app (choosing an existing product or a new one named after the directory); call again with the returned requestId once approved and the telora-daemon connect flow wires the directory (git repo, daemon.json, repo hooks) and starts the daemon -- then the user restarts Claude Code here. It never mints a daemon credential from the session's own tracker. In a directory already bound, it (re)starts the daemon for it.
telora_session_state1 of 1 parameter describedSession bootstrap: returns Telora connection state in a single payload (connected URL, organization, product, active focus + role, pending delivery counts). activeFocus also carries an operator-stop indicator (held: true + operatorStoppedAt timestamp) so a stop is confirmable without reading the daemon log. Used by the SessionStart hook to surface state at chat open. Optional productId; auto-resolved when the organization has a single product. NOTE: branch state (commits ahead of integration) is NOT included in this response -- the server has no git access. For branch state, run `telora-daemon session-state --json` in the repo root.
telora_organization5 of 5 parameters describedOrganization settings: the org's per-pass DEFAULT models and efforts (coding, review). Models: the model a focus pass runs when the focus pins none. The harness equivalent of the dashboard's Settings > Default Pass Models pickers, and it accepts exactly what they save: a selectable hosted model id (claude-fable-5-2, claude-fable-5-1, claude-opus-5-5, claude-opus-5, claude-opus-4-8, claude-sonnet-5-5, claude-sonnet-5, claude-haiku-4-5-20251001, gpt-6-astra, gpt-6-sol, gpt-6-luna, gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna), or an empty string to clear (inherit). Legacy ids that are no longer selectable (e.g. gpt-5-codex, gpt-5, o3-mini, claude-fable-5) are refused; an org already pinned to one still reads it back. Model precedence: a focus pin > an org-authored workflow stage model > this org default > the builtin workflow stage model. Engine: the daemon runs a model on the engine it implies (claude-* on Claude Code, gpt-* on Codex) when the model's layer outranks the pass's engine choice, on one ladder: focus > org-authored workflow stage > org default > product default engine (daemon.json on the daemon host) > builtin workflow stage > global default. So an org default gpt-* model runs on Codex even when the org or product default engine is Claude Code (and an org default claude-* model runs on Claude Code on a Codex-harness product), but a focus engine override or an org-authored stage engine outranks it; a model the chosen engine cannot run is dropped and that engine runs its own default. Per-pass DEFAULT effort (coding, review): one of low, medium, high, xhigh, max, or an empty string / null to clear (engine default -- no effort flag). Effort precedence: a focus's pipelineConfig.efforts > this org default > the engine default. Each engine runs the nearest level it accepts: Codex has no max, so max runs as xhigh. Actions: get_pass_models (read the current defaults -- models AND efforts), set_pass_models (write models; only the passes you provide change; an empty string clears -> inherit), set_pass_efforts (write efforts via codingEffort/reviewEffort; only the passes you provide change; an empty string or null clears -> engine default).
telora_discover3 of 3 parameters describedLoad additional Telora tool domains on demand. At startup only core tools (product, focus, delivery, issue) are available. Use this to load domains for: agent, role, escalation, AI, blocked, help, wiki, page, knowledge, search, tree, navigation, split, merge, archive, audit, context, workflow, state machine, transition, guard, policy, lifecycle, stage, report, stakeholder, weekly, digest, summary, progress, template, cadence, playbook, pattern, instantiate, repeatable, deployment, profile, CI/CD, environment, infrastructure, deploy, risk, register, mitigation, dependency, communication, impact, likelihood, cascade, reality tree, CRT, FRT, Goldratt, root cause, planning, graph, injection, node, edge, undesired effect, UDE, desired effect, leverage, drift, lineage, traverse, focus-scoped, overlay, dump, causal, thinking processes, loop, document, question, answer, persona, intent, differential, signal, variance, reality metrics, metrics, projection, telemetry, evaluate, brief, wire A, wire B, receiver, oee, workstation oee, workstation, prd, milestone, MVP, app in one run, one run, one go, whole app, entire app, end-to-end, build an app, fully working app, prerequisite tree, PRT, goal, obstacle, overcome, clearance predicate, vision, pathway, frontier, materialize, contract, agreement, counterparty, version, diff, markdown, local model, registry, LLM, codex, endpoint, base_url, wire_api, local, model server, ollama, self-hosted, per-pass, integration, ticketing, tracker, linear, connection, data source, api key, backlog, TIN-, external, look at issue, client asks, provenance, informed_by, reference, cite, issue, initiative, project, cycle, label, sprint, review the backlog, source of intent, provider, consumer, versioned, boundary, interrelation, cross-product, nexus, public face, blast radius, consume, service definition, rationale, exposed surface, stability, audience, what to expose, register decision, intent layer, starter, starter repo, library, scaffold, genesis, boilerplate, app shell, clone, polished, ux floor, greenfield, from zero, archetype, factory, request, waiting on, requested by, lateral, peer product, return, pending return, acknowledge, code observation, quality, maturity, artifact, distillate, judgment, review, hotspot, station, deposit, code health, instrumentation, migration, migration number, schema change, claim, ordinal, collision, numbering, supabase migration, drizzle, journal, next free number, tower, run order, queue, depends on, excludes, parallel foci, sequence foci, arm several foci, concurrency, tower decision, supervisor decision, review loop, bounded criterion, radar, blip, feedback, feature request, change request, problem, idea, defect, transcript, email, meeting notes, feedback document, observed demand, evidence, quote, glossary, mis-transcription, customer feedback, not yet intent, theme, force rank, radar order, waiting, commitment, promise, overdue, release, releases, release notes, changelog, ship, shipped, cut, cut a release, staging, production, what shipped, release candidates, publish, publish release notes, edit release note, broken links, check links. Available domains: agent: AI agent roles and human-AI escalation management (1 tool, 5 actions) wiki: AI-managed product wiki: ranked search, page CRUD, navigation tree, split, merge, archive, and audit (9 tools) workflows: Workflow state machines with transition guards and policy automation (1 tool, 5 actions) reports: Stakeholder reports and templates: AI-generated digests, risk assessments, custom report types (2 tools, 8 actions) playbooks: Playbook templates for repeatable execution patterns (1 tool, 4 actions) profiles: Deployment profiles for CI/CD and environment configuration (1 tool, 5 actions) risks: Risk register: risk CRUD, risk-to-risk links, focus links, external dependency tracking, communication log (3 tools, 17 actions) reality_trees: Reality trees (Goldratt Thinking Processes): focus-scoped causal C…
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.