All servers / Browser / Chrome DevTools MCP server
Chrome DevTools MCP server
The power user's browser MCP — connects to your real Chrome/Edge/Brave with full DevTools capabilities
Tested by ConnectATDScore 86 out of 100. Tested 2026-11-19 on package version 5.5.2. Package: cdp-mcp-server · source
Score breakdown
| Runs | 18 / 20 | |
|---|---|---|
| Tool descriptions | 20 / 20 | |
| Package safety | 20 / 20 | |
| Maintenance | 20 / 20 | |
| Adoption | 8 / 20 |
What we saw
- Starts and answers the MCP handshake: yes (3.1 seconds including download)
- Tools offered: 15
- Last published: 2026-11-17
- Weekly downloads (npm): 622
- License: MIT · Direct dependencies: 2
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.
pingno parametersHealth check. Returns server version and connection status.
statusno parametersServer status: connection, config, sessions.
tabs12 of 12 parameters describedTab management. The server auto-connects to your browser — no need to call browser.connect() first. Just use any tool and it works. Default cleanup is 'detach' — tabs stay open after your session ends. You don't need to set cleanupStrategy unless you want tabs auto-closed. Sessions are automatically pinned to the Chrome profile where they first create or claim a tab. New tabs always open in the session's profile, even if you manually switch profiles in Chrome. Use tabs.new with profile:'Name' to override, or tabs.set_profile to change the session's default profile. Use showAll: true to see ALL browser tabs, not just ones you created this session. Operations: - list: List all open browser tabs with their IDs, URLs, and titles (optional: urlFilter — substring filter, format — 'text' default | 'json' for parseable [{id,url,title,profile}]) - find: Search tabs by title or URL substring (requires: query; optional: format — 'json' for parseable output) - new: Open a new tab (optional: url — defaults to about:blank, activate — bring to foreground, default: false, profile — create in specific Chrome profile by name/email/directory) - close: Close a specific tab (requires: tabId) - activate: Bring a tab to the foreground (requires: tabId) - info: Get detailed info about a tab including URL, title, and connection status (requires: tabId) - set_profile: Pin session to a specific Chrome profile (requires: profile — name/email/directory)
page46 of 46 parameters describedPage operations. Take a snapshot before interacting — it gives you uid refs for elements. For slow pages (D365, SharePoint), the default 60s timeout should work. Use timeout parameter for even slower pages. The snapshot is optimized for LLM consumption — decorative elements are filtered out. Operations: - goto: Navigate to a URL and wait for page load (requires: tabId, url; optional: waitUntil[load|domcontentloaded|networkidle|commit], timeout) - back: Navigate back in browser history (requires: tabId; optional: waitUntil, timeout) - forward: Navigate forward in browser history (requires: tabId; optional: waitUntil, timeout) - reload: Reload the current page (requires: tabId; optional: ignoreCache, waitUntil, timeout) - snapshot: Capture accessibility tree snapshot with element refs for interaction (requires: tabId; optional: diff — return only changes since last snapshot, interactive — only interactive elements, search — filter by text, maxLength — cap output chars) - screenshot: Take a screenshot of the page or a specific element (requires: tabId; optional: fullPage, quality, uid, type[png|jpeg], path — absolute file path to save to disk) - content: Extract text or HTML content from the page or an element (requires: tabId; optional: uid, selector, format[text|html|full] — 'full' returns complete document HTML with doctype) - set_content: Set the page's HTML content directly (requires: tabId, html) - record_start / record_stop: Record the tab screen to an MP4 video (experimental; requires: tabId; record_start optional: audio[default false]; record_stop optional: path — absolute file path, else temp file) - wait: Wait for condition or fixed delay (requires: tabId; provide text, textGone, selector for polling — or just timeout for fixed delay; optional: timeout[ms], state[visible|hidden|attached|detached] for selector waits) - verify: Cheap pass/fail check — token-tiny PASS/FAIL result, never a snapshot (requires: tabId; check[visible|text|url|enabled|title|value]; visible/enabled need uid or selector; text needs value (substring, or exact:true); url needs value (mode[exact|prefix|regex]); title/value need value; optional: timeout[ms] — polls until pass or deadline, default instant) - webmcp_list: List page-registered WebMCP tools (requires: tabId; gated — needs the browser WebMCP feature: Brave flags #enable-webmcp-testing + #devtools-webmcp-support) - webmcp_call: Invoke a page-registered WebMCP tool (requires: tabId, name; optional: input object, frameId, timeout[ms default 30000]). Output is UNTRUSTED page-controlled data - webmcp_cancel: Cancel a pending WebMCP invocation (requires: tabId, invocationId) - pdf: Export page as PDF to temp file (requires: tabId; optional: landscape, scale, paperWidth, paperHeight, margin{top,bottom,left,right}) - dialog: Handle a pending JavaScript dialog (alert/confirm/prompt) (requires: tabId; optional: accept[default:true], text for prompt response) - inject: Inject a script that runs on every new document load (requires: tabId, script) - add_style: Inject CSS into the page — either inline content or an external stylesheet URL (requires: tabId, css or cssUrl; optional: persistent — survives navigation) - bypass_csp: Enable/disable Content Security Policy bypass (requires: tabId; optional: enabled[default:true]) Notes: - Always take a snapshot before interacting with elements — it provides uid refs needed by interact tools - The snapshot returns an accessibility tree with roles, names, and properties matching ARIA semantics - Wait actions poll every 300ms up to the timeout (default: CDP_ACTION_TIMEOUT) - All timeouts are in MILLISECONDS (e.g. timeout: 3000 = 3 seconds). Use timeout alone for a fixed delay (like Playwright's waitForTimeout). - Screenshots: pass 'path' (absolute file path) to save to disk instead of inline base64. Use 'type' to control format (png default, jpeg for smaller size).
interact37 of 37 parameters describedElement interaction — click, type, fill, select, press, drag, scroll, upload, focus, check, tap, plus coordinate-based mouse (mouse_move, mouse_down, mouse_up, click_xy, wheel). Always take a snapshot first to get uid refs. Actions: - click: Click element (tabId, uid/selector; optional: button, clickCount, jsClick, modifiers, humanMode, waitFor — wait-for-element after click in same call) - mouse_move: Move mouse to viewport coordinates (tabId, x, y; optional: humanMode — bezier mouse path) - mouse_down: Press mouse button at viewport coordinates (tabId, x, y; optional: button, clickCount, modifiers) — press-hold workflows - mouse_up: Release mouse button at viewport coordinates (tabId, x, y; optional: button, clickCount, modifiers) - click_xy: Click at viewport coordinates without an element (tabId, x, y; optional: button, clickCount, modifiers, humanMode) — canvas/map/virtualized UIs; hit-target is the caller’s responsibility - wheel: Dispatch mouse wheel (tabId, deltaX and/or deltaY, optional: x, y, modifiers) — positive deltaY scrolls down - hover: Hover element (tabId, uid/selector) - type: Type text (tabId, text, uid/selector; optional: clear, submit, delay, charDelay, wordDelay, typoRate) - fill: Fill multiple fields [{uid, value, type}] (tabId, fields) - select: Select dropdown option (tabId, value, uid/selector) - press: Press key (tabId, key; optional: modifiers) - drag: Drag between elements (tabId, sourceUid/sourceSelector, targetUid/targetSelector) - scroll: Scroll page/element (tabId; optional: direction, amount, x, y, uid/selector) - upload: Upload files (tabId, files; optional: uid/selector — without uid monitors for file chooser dialogs) - focus: Focus element (tabId, uid/selector) - check: Set checkbox state (tabId, checked, uid/selector) - tap: Touch tap element (tabId, uid/selector) Use uid (from snapshot) or a locator (selector). UIDs preferred. Locators: Plain CSS, or a Playwright-style engine: css=, //xpath (or xpath=), text= ("quoted" = exact match), role= (aria-lite), label=, placeholder=, testid=, alt=, title=. Append " >> nth=N", " >> first" or " >> last" when multiple elements match. Open shadow roots are pierced (xpath is document-root only). Set humanMode:true for realistic mouse paths.
execute10 of 10 parameters describedRun JavaScript on the page. Use this when standard tools can't handle a specific interaction. For clicking elements that don't respond to CDP clicks (rare): execute({ action: 'call', uid: 5, function: '(el) => el.click()' }) Operations: - eval: Evaluate a JavaScript expression and return its value (requires: tabId, expression) - script: Execute an async function body wrapped in an IIFE — use for multi-step JS logic (requires: tabId, code) - call: Call a JavaScript function with a specific page element as its argument (requires: tabId, function — e.g. '(el) => el.textContent', uid or selector) Notes: - eval returns the expression's value directly (must be JSON-serializable) - script wraps your code in: (async () => { <your code> })() — use 'return' to send a value back - call resolves the element first, then passes it as the first argument to your function
observe19 of 19 parameters describedUse this to debug page behavior — check console errors, inspect API calls, or measure performance. Monitoring starts automatically when queried. Operations: - console: Retrieve captured console messages (requires: tabId; optional: level[all|error|warning|log|info|debug], last — return only last N entries, clear — clear after returning) - network: List captured network requests with URLs, methods, status codes, and timing (requires: tabId; optional: filter — URL substring, types — resource type filter array, last, clear) - request: Get the full request and response body for a specific network request (requires: tabId; provide either requestId — from network listing — or urlFilter — substring of the URL; urlFilter resolves to the most recent matching request) - performance: Collect page performance metrics including DOM size, JS heap, layout counts, and paint timing (requires: tabId) - perf_start: Begin recording page rendering performance via CDP Tracing (LCP candidates + layout shifts + interaction latency). Returns immediately. Call observe perf_stop on the same tab to finish and extract insights. Capture quality depends on the page rendering during the recording; per-frame paint-metric emission to the trace is nondeterministic between runs on some browsers — a zero-class result can succeed on a retry. Interaction timing (EventTiming/INP) requires categories: ["v8"] AND real input during the window (requires: tabId; optional: categories — ["v8"] enables the v8+devtools.timeline set; other values are rejected; optional: keepRaw — also stream every raw trace event to a .jsonl temp file, off by default) - perf_stop: Stop the recording started by perf_start on this tab and aggregate it into token-capped insights: LCP (last candidate per frame), CLS (final cumulative score, input shifts excluded), and interaction latency (EventTiming top-3 by duration when v8 + input captured it). If the session died mid-recording, the stop reports the recording as lost (requires: tabId) - heap_dump: Stream a full .heapsnapshot of this tab's renderer heap to a session temp file (chunks land in a *.partial file renamed on completion; discoverable by cleanup clean_temp, never auto-deleted). Dumps can carry auth-state material from page JS — opt-in required. Review offline with heap_summary (requires: tabId; confirm: true) - heap_summary: Offline parse + summary of a .heapsnapshot file — node counts, top constructors by SELF size, detachedness rollup, bounded GC-root paths. Self-size semantics (no dominator tree; "retained" is not claimed). Requires a live browser connection like any tool, but parses the file in-process (requires: path — the heap_dump-returned file; optional: top, hintDepth, hintBudget) - heap_diff: Offline diff of TWO .heapsnapshot files (A → B) by SELF size. Identity = constructor (type+name) with an EXACT self-size breakdown (never node ids); classes NEW / GONE / GROWN / SHRUNK, node/string deltas, plus a separate ≈ byte-bucket rollup. No dominator tree — "retained" is not claimed (requires: pathA, pathB — the heap_dump-returned files; optional: top) - memory: DOM/JS memory counters (documents, nodes, JS event listeners) before/after a forced GC sweep via Memory.prepareForLeakDetection, plus leak-detection counters when supported (requires: tabId) - downloads: List tracked file downloads with progress info (requires: tabId; optional: last, clear) - har: Export captured network requests as HAR 1.2 JSON (requires: tabId) Network Resource Types: xhr, fetch, document, script, stylesheet, image, font, media, websocket, other Notes: - Console and network monitoring starts automatically when first queried — no explicit enable needed - Popups/new windows opened by pages are auto-detected and logged to the opener tab's console as [popup] entries - Use 'clear: true' to reset captured data between test iterations
emulate28 of 28 parameters describedUse this to test responsive design, mobile views, or simulate slow networks. Set any combination of properties in one call. Pass 'reset: true' to clear all overrides. Operations (set any combination in a single call): - device: Apply a device preset (optional: device name or list; overrides viewport+UA unless explicitly given) - viewport: Set viewport dimensions (optional: {width, height, deviceScaleFactor, mobile, touch, landscape}) - colorScheme: Emulate preferred color scheme (optional: dark|light|auto) - userAgent: Override the browser user agent string (optional: string) - geolocation: Spoof geolocation (optional: {latitude, longitude, accuracy, altitude}) - cpuThrottle: Throttle CPU speed — 1 = normal, 4 = 4x slower (optional: number) - timezone: Override timezone (optional: string, e.g. 'America/New_York') - locale: Override locale (optional: string, e.g. 'fr-FR') - visionDeficiency: Simulate vision impairment (optional: none|protanopia|deuteranopia|tritanopia|achromatopsia|blurredVision) - autoDarkMode: Force automatic dark mode (optional: boolean) - idle: Emulate idle/locked screen state (optional: active|locked) - devicePosture: Foldable device posture override (optional: continuous|folded; Chrome 153+) - cpuPerformance: CPU performance tier override (optional: low|mid|high|ultra; Chrome 153/154+) - networkCondition: Throttle network speed (optional: offline|slow3g|fast3g|slow4g|fast4g|none; uses Network.emulateNetworkConditionsByRule on Chrome 152+, legacy transport otherwise; cleared by reset) - networkRules: Per-request throttle rules (optional: array of {urlPattern? (empty=all, absolute https URLPattern otherwise), latency, downloadThroughput, uploadThroughput, packetLoss?}; first-match wins; Chrome 152+; cleared by reset) - networkRulesList: List the active per-request rules with indices + evaluation order (optional: true; read-back only, no CDP mutation; Chrome 152+) - networkRulesAppend: Append ONE rule (same shape as a networkRules entry) and re-push the whole list in one ByRule call (optional: object; Chrome 152+) - networkRulesRemove: Remove the rule at this index and re-push the remainder in one ByRule call (optional: number; Chrome 152+) - overrideNetworkState: Spoof navigator.onLine/connection (optional: boolean; true=offline, false=online; Network.overrideNetworkState, Chrome 152+; cleared by reset) - inputVisualization: Show click markers on the page (optional: boolean; Page.setInputVisualization, Chrome 153+; cleared by reset) - ignoreSSL: Bypass SSL certificate errors (optional: boolean) - blockUrls: Block requests matching URL patterns (optional: string array) - extraHeaders: Set extra HTTP headers on all requests (optional: object) - reset: Clear ALL emulation overrides (optional: true) Network Presets: offline (no connection), slow3g (2s latency, 50KB/s), fast3g (562ms, 180KB/s), slow4g (150ms, 400KB/s), fast4g (50ms, 1.5MB/s), none (remove throttling)
storage23 of 23 parameters describedUse this to manage cookies and browser storage — inspect auth tokens, set test cookies, read/write localStorage or sessionStorage, or clear site data for a clean state. Operations: - set_cookie: Set a cookie (requires: tabId, name, value; optional: domain, path, secure, httpOnly, sameSite[None|Lax|Strict], expires — epoch seconds, partitionKey — CHIPS partitioning: origin string (Playwright sourceOrigin shorthand) or {topLevelSite, hasCrossSiteAncestor}; partitioned cookies are set Secure automatically; cleared). - delete_cookies: Delete specific cookies by name (requires: tabId, name; optional: url, domain, path, partitionKey — partition-scoped delete) - get_cookies: Retrieve cookies (requires: tabId; optional: urls — filter by URLs; partitioned cookies show | partitioned: <topLevelSite>) - clear_cookies: Clear all cookies for the browser profile (requires: tabId) - clear_data: Clear browser storage for an origin (requires: tabId; optional: origin, types — comma-separated: 'cookies,local_storage,indexeddb,cache_storage' or 'all') - quota: Check storage quota usage for the current page's origin (requires: tabId) - get_storage: Read localStorage/sessionStorage. Without 'key' returns the whole object as JSON (requires: tabId; optional: storageType[local|session, default local], key) - set_storage: Write a localStorage/sessionStorage key (requires: tabId, key, value; optional: storageType[local|session, default local]) - delete_storage: Remove a key (or clear all when omitted) from localStorage/sessionStorage (requires: tabId; optional: storageType[local|session, default local], key) - export_state: One-shot auth-state export — all cookies + current origin's localStorage as a Playwright-compatible JSON object (requires: tabId; optional: includeSessionStorage; partitioned cookies export with their partitionKey). Feed the returned object into import_state - import_state: Restore a storage state — sets cookies browser-wide and applies localStorage for the CURRENT page origin (other origins reported as skipped — storage is per-origin; navigate there and re-import) (requires: tabId, state; optional: clearFirst — clear all cookies first)
intercept14 of 14 parameters describedRequest interception requires careful use. Paused requests that aren't handled within 30s will be auto-continued to prevent page hangs. Always call intercept.disable() when done. Operations: - enable: Start intercepting requests matching URL patterns (requires: tabId; optional: patterns — array of URL glob patterns, e.g. ['*.api.example.com/*']) - disable: Stop all request interception (requires: tabId) - continue: Resume a paused request, optionally modifying it (requires: tabId, requestId; optional: url, method, headers, postData) - fulfill: Respond to a paused request with a custom/mocked response (requires: tabId, requestId; optional: status, body, headers) - fail: Abort a paused request with a network error (requires: tabId, requestId; optional: reason) - list: List all currently paused/intercepted requests (requires: tabId) Failure Reasons: Failed, Aborted, TimedOut, AccessDenied, ConnectionClosed, ConnectionReset, ConnectionRefused, ConnectionAborted, ConnectionFailed, NameNotResolved, InternetDisconnected, AddressUnreachable, BlockedByClient, BlockedByResponse Notes: - Call 'enable' first with URL patterns to start intercepting - Intercepted requests are paused until you call continue, fulfill, or fail - Each intercepted request has a unique requestId shown in the 'list' output
cleanup6 of 6 parameters describedSession and connection management. Use 'status' to check server health. WARNING: 'reset' terminates ALL sessions. Only use if something is fundamentally broken. Operations: - disconnect_tab: Disconnect from a specific tab session without closing the tab (requires: tabId) - disconnect_all: Disconnect all active tab sessions owned by this session (no parameters) - clean_temp: Delete temporary files (screenshots, PDFs, response dumps) created by this session (no parameters) - status: Show current server status — active sessions, temp file count, connection state (no parameters) - list_sessions: List all active agent sessions with their TTL, idle time, cleanup strategy, and associated tabs (no parameters) - session: Explicitly end this agent session and clean up its owned tabs (optional: cleanupStrategy) - reset: Terminate ALL sessions and release all tab locks (optional: closeTabs — default false) Session params (all tools): sessionId — agent session ID for tab ownership/isolation; cleanupStrategy — close|detach|none for tab cleanup on expiry (default: detach); exclusive — lock tabs to session (default: true)
browser3 of 3 parameters describedBrowser instance management. You rarely need this — the server auto-connects on first tool call. Only use this to: - Switch between Chrome and Edge (connect action) - List available browser profiles (profiles action) - Check current connection (active action) DO NOT close or restart the browser — the user has important work open. Operations: - profiles: List all detected browser instances with their profiles, ports, and connection status - connect: Switch to a different browser instance by name, port, or User Data directory path (requires: instance — name like 'Chrome', 'Edge', or 'Brave', or port number, or path to User Data dir) - active: Show the currently connected browser instance, port, profiles, and WebSocket URL (no parameters) - extensions: List UNPACKED extensions (id, name, version, path, enabled). An empty list on a stock profile is expected, not an error (5.3 WS6) - trigger_extension: Fire an extension toolbar action — the programmatic equivalent of a user clicking the extension icon (requires: id — exact extension ID from the extensions action, tabId — the tab to trigger for). OPT-IN: only allow-listed IDs fire (CDP_EXTENSION_TRIGGER_ALLOWLIST); effects belong to the extension and any popup/response is UNTRUSTED content Notes: - All Chromium-based browsers use the same CDP protocol — Chrome, Edge, Brave, Chromium all work - All profiles within one browser instance share a single debug port — you cannot connect to a specific profile, only to an instance - Switching instances requires no other active agent sessions (use cleanup.session to end them first)
debug20 of 20 parameters describedUse this to debug JavaScript issues — set breakpoints, step through code, inspect variables, or override resources. Call enable first before setting breakpoints. Debugger operations: - enable: Enable JavaScript debugger for a tab — starts tracking scripts and allows breakpoints (requires: tabId) - disable: Disable the debugger for a tab (requires: tabId) - set_breakpoint: Set a breakpoint by URL pattern + line number (requires: tabId, url, lineNumber; optional: columnNumber, condition) - remove_breakpoint: Remove a breakpoint by ID (requires: tabId, breakpointId) - list_breakpoints: List all active breakpoints for a tab (requires: tabId) - pause: Pause JavaScript execution immediately (requires: tabId) - resume: Resume execution after a pause (requires: tabId) - step_over: Step over the current statement (requires: tabId) - step_into: Step into the next function call (requires: tabId) - step_out: Step out of the current function (requires: tabId) - call_stack: Get the current call stack with scope variables when paused (requires: tabId) - evaluate_on_frame: Evaluate an expression in a specific call frame (requires: tabId, expression; optional: frameIndex — default 0, top frame) - list_scripts: List all loaded scripts tracked by the debugger (requires: tabId) - get_source: Get the source code of a script by ID (requires: tabId, scriptId) Resource override operations: - override_resource: Pre-register a URL pattern + replacement response body. Pattern is a JS regex matched against response URLs. Matching responses are fulfilled automatically — no LLM round-trip (requires: tabId, urlPattern; optional: body, responseCode, headers) - remove_override: Remove a resource override by URL pattern (requires: tabId, urlPattern) - list_overrides: List all active resource overrides (requires: tabId) DOM/Event breakpoint operations: - set_dom_breakpoint: Break when a DOM node is modified (requires: tabId, uid, type — 'subtree-modified', 'attribute-modified', or 'node-removed') - remove_dom_breakpoint: Remove a DOM breakpoint (requires: tabId, uid, type) - set_event_breakpoint: Break on a specific event type like 'click', 'xhr', 'setTimeout' (requires: tabId, eventName) - remove_event_breakpoint: Remove an event breakpoint (requires: tabId, eventName) Notes: - Call 'enable' before setting breakpoints or listing scripts - When debugger pauses, ALL other tool calls on that tab are blocked until you resume/step - Auto-resume fires after 30s (CDP_DEBUGGER_TIMEOUT) to prevent permanently frozen tabs - Resource overrides work independently of the debugger — no need to call 'enable' first - Resource overrides coexist with request interception (intercept tool). Both can be active simultaneously - Performance: while overrides are active, ALL responses are routed through the handler for regex matching. Remove overrides when done to avoid unnecessary overhead on resource-heavy pages Session params (all tools): sessionId, cleanupStrategy, exclusive
form5 of 5 parameters describedSmart form filling — handles text, combobox, checkbox, radio, select, date, file, country-code, location in one call. Auto-detects field type from a11y role. For comboboxes, types char-by-char and waits for dropdown options. Actions: fill (set values), read (get current values), clear (reset fields). Field types (auto-detected): textbox/searchbox/textarea → text, combobox → char-by-char + dropdown, checkbox/switch → toggle, radio → click, listbox/select → native or custom dropdown, country-code → normalizes +91/India/IN, location → slow type + API autocomplete. Example: form({ action: "fill", tabId: "...", fields: [{ uid: 5, value: "John" }, { uid: 8, value: "India", type: "location" }] }) Always take a snapshot first to get uid refs.
webauthn27 of 27 parameters describedUse this to test passkey/WebAuthn login flows with virtual authenticators — no real security key needed. Create a virtual authenticator on a tab, then drive the page normally: the browser intercepts navigator.credentials.create/get and the virtual authenticator auto-responds. IMPORTANT: virtual authenticators live in browser memory only — nothing touches the user's real keyring. They die when the tab closes, the browser exits, or this server disconnects; they SURVIVE in-tab session re-attaches (process swaps) orphaned from the registry. Create the authenticator BEFORE the page runs its WebAuthn calls. The passkey flow's success signal is the PAGE's state afterwards — verify with page snapshots, not webauthn return values. Caveat: pages using conditional mediation (page-load autofill WebAuthn) can starve the site's own credentials.create — reload first, or pre-provision with add_credential. CAVEAT 2: if a password-manager extension with passkey interception is active (Bitwarden, etc.), it proxies ALL page WebAuthn requests at the browser level and the virtual authenticator is bypassed — assertions time out and creates go to the real vault. Disable the extension's passkey support when testing virtual-authenticator flows. Operations: - add_authenticator: Create a virtual authenticator (requires: tabId; optional: protocol[ctap2|u2f], transport[internal|usb|nfc|ble|cable|hybrid|smart-card], hasResidentKey, hasUserVerification, isUserVerified, automaticPresenceSimulation, ctap2Version[ctap2_0|ctap2_1|ctap2_2], extensions{prf,largeBlob,credBlob,minPinLength,hmacSecret,cmtgKey}). Defaults: ctap2/internal/resident-key/UV-verified/auto-presence — a platform passkey that approves everything. Re-enables the virtual environment automatically. - remove_authenticator: Remove one (requires: authenticatorId) - list_authenticators: List authenticators created by this server (no params) - add_credential: Inject a pre-made credential (requires: authenticatorId, credentialId[b64], privateKey[b64 PKCS#8], rpId; optional: userHandle[b64], isResidentCredential, signCount, largeBlob[b64], backupEligibility, backupState, userName, userDisplayName) - list_credentials: List credentials stored on an authenticator (requires: authenticatorId; optional: fullIds — print untruncated credential IDs, needed for remove_credential / set_credential_properties) - remove_credential: Remove one credential (requires: authenticatorId, credentialId) - clear_credentials: Remove all credentials (requires: authenticatorId) - set_user_verified: Make user verification succeed/fail (requires: authenticatorId; optional: isUserVerified, default true) - set_presence_simulation: Auto-approve or hang user-presence tests (requires: authenticatorId; optional: enabled, default true) - set_override_bits: Negative testing — sabotage the next authenticator response (requires: authenticatorId; optional: isBogusSignature, isBadUV, isBadUP — zeros the signature / UV flag / UP flag) - set_credential_properties: Mutate a stored credential — backup flags and/or the signature counter (requires: authenticatorId, credentialId; optional: backupEligibility, backupState, signCount — signCount -1 REMOVES the counter, useful for testing signCount-monotonicity handling in relying parties). Credential lifecycle events (added/updated/asserted/deleted) surface in list_credentials as "last:" lines - reset: Remove ALL authenticators on a tab + disable the environment (requires: tabId)
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.