Bragproof

← Reports

@avallon-labs/mcp

Starts with no setup68.3.0npm

Installed, started with no arguments, no environment and no input, and listed its tools. Run 2026-11-16.

Needs to start: nothing
Returns a result when called with no setup: 0 of 5 tools tried
Contacts while installing and starting: nothing

200tools listed
0hosts contacted
0files written
0.3 sto start
95dependencies

Its tool definitions changed since version 68.2.0. If you approved the earlier ones, read them again.

What it contacted

Nothing besides the npm registry it was installed from. Watched from outside its sandbox while it was installed and started with no input. Its tools may still contact things when called.

Starting it

npx -y @avallon-labs/mcp@68.3.0

Installing a package runs its code. Use a sandbox for anything you have not read.

Calling its tools

5 tools that take no input and look read-only were called once with nothing. 0 returned a result.

listWorkerModelsNo usable answer tool reported an error: Avallon API error 401: missing authentication credentials. Provide either Authorization header (JWT) or x-api-key header is not authoriz
getProfileNo usable answer tool reported an error: Avallon API error 401: missing authentication credentials. Provide either Authorization header (JWT) or x-api-key header is not authoriz
listTeamMembersNo usable answer tool reported an error: Avallon API error 401: missing authentication credentials. Provide either Authorization header (JWT) or x-api-key header is not authoriz
getTenantSettingsNo usable answer tool reported an error: Avallon API error 401: missing authentication credentials. Provide either Authorization header (JWT) or x-api-key header is not authoriz
listEvaluatorsNo usable answer tool reported an error: Avallon API error 401: missing authentication credentials. Provide either Authorization header (JWT) or x-api-key header is not authoriz

During the calls (not necessarily because of them) it first contacted: api.avallon.ai.

Only tools whose names read like looking something up, with no required input, are tried, five at most. Tools that need input, and anything that sounds like it changes or sends something, are not called.

What its tools tell an agent

In the tool descriptions, nothing that talks about the user, other tools or private files, or uses hidden characters.

Found by simple pattern rules in each tool's description (up to 3,000 characters), quoted as written. Parameter descriptions are not scanned. A clean result is not a safety review.

Tools

listVoiceAgents queryParams
List all voice agents for your tenant.

**Filtering:**

Filter voice agents by name using the `agent_name` query parameter (exact match).

**Response:**

Returns an array of all voice agents with their configurations and a total count.

For compatibility this endpoint is also available using the following (deprecated) paths:
`GET /v1/agents`
createVoiceAgent bodyParams
Create a new voice agent with the specified configuration.

**Required fields:**

Provide `engine`: a `pipeline` of STT, LLM and TTS models, or a `speech_to_speech` model such as `gpt-live-1`. The flat `llm_model`, `tts_provider`, `tts_model`, `tts_voice_id` and `stt_model` fields are deprecated; they still create a pipeline agent but cannot be combined with `engine`.

All other fields are required except `phone_number_id` and `transfer_phone_number`.

**Response:**

Returns the created voice agent with its ID and configuration.

For compatibility this endpoint is also available using the following (deprecated) paths:
`POST /v1/agents`
getVoiceAgent pathParams
Retrieve a single voice agent by its ID.

For compatibility this endpoint is also available using the following (deprecated) paths:
`GET /v1/agents/{id}`
updateVoiceAgent pathParams, bodyParams
Update a voice agent. Only the fields you send change; omitted fields keep their current value.

To change the processor assignment, `call_processor_id` and `call_processor_type` must be provided together. Set `call_processor_id` to `null` to detach (`call_processor_type` is still required).

Set `engine` to replace the whole model setup: `pipeline` (STT, LLM and TTS models plus optional fallbacks) or `speech_to_speech` (`gpt-live-1` and a voice). For `gpt-live-1`, `prompt` is the reasoner prompt and `speaker_instructions` the speaker prompt.

The flat model fields (`llm_model`, `tts_provider`, `tts_model`, `tts_voice_id`, `stt_model` and their `fallback_*` counterparts) are deprecated and only work on pipeline agents. They cannot be combined with `engine`. They are validated against the resulting configuration, so a single field such as `tts_voice_id` can be sent on its own.

Set `voice_mail_detection` to `true` to have the agent listen for a voicemail beep tone. Detection only runs on `OUTBOUND` agents; enabling it on an `INBOUND` agent is stored but has no effect.

Set `max_call_duration_seconds` to change how long a call may run before the agent ends it (default 1800).

Set `background_sounds` to `disabled` to turn off the ambient office background sounds played during calls.

Set `noise_cancellation` to choose the noise cancellation mode applied to the caller audio.

Set `tool_ids` to replace the agent's attached tool set. This is the supported way to manage tool attachments; the dedicated add and remove endpoints are deprecated.

Set `evaluator_slug` to score the agent's calls with that evaluator. Calls only use it once the new version is promoted, because a call is evaluated with the version it ran on.

All changes in one request create a single new voice agent config version. The live and test releases keep pointing at their current versions until the new version is promoted.
deleteVoiceAgent pathParams
Delete a voice agent. The agent is hidden from all reads; its call and job history is preserved.
195 more tools
listVoiceAgentVersions pathParams, queryParams
List all config versions for a voice agent, ordered by version (newest first by default).
getVoiceAgentVersion pathParams
Retrieve a single voice agent config version, including the agent's identity, the config snapshot at that version, and version-specific metadata.
promoteVoiceAgentVersion pathParams, bodyParams
Select a voice agent configuration version for the live or test environment.
getVoiceAgentVersionChangeSummary pathParams
Returns an AI-generated summary of what changed between this version and the previous one. Initial version returns an empty string since there is no predecessor.
getVoiceAgentPlaceholders pathParams
Retrieve the available placeholder variables defined in a voice agent's prompt.

For compatibility this endpoint is also available using the following (deprecated) paths:
`GET /v1/agents/{id}/prompt`
getVoiceAgentPrompt pathParams, queryParams
Retrieve the prompt for a voice agent. Pass `version` to fetch the prompt snapshot at a specific config version; omit it to fetch the current prompt. Internal use only.
updateVoiceAgentPrompt pathParams, bodyParams
Update the prompt for a voice agent. Internal use only.

**Deprecated:** Use `PATCH /v1/voice-agents/{voiceAgentId}` instead.
updateVoiceAgentModels pathParams, bodyParams
Update the LLM, TTS, and STT model configuration for a voice agent.

**Deprecated:** Use `PATCH /v1/voice-agents/{voiceAgentId}` instead.

For compatibility this endpoint is also available using the following (deprecated) paths:
`PATCH /v1/agents/{id}/models`
updateCallControls pathParams, bodyParams
Update call control settings for a voice agent.

Configure whether the voice agent supports dial pad input, can end calls, and set a transfer phone number.

**Deprecated:** Use `PATCH /v1/voice-agents/{voiceAgentId}` instead.

For compatibility this endpoint is also available using the following (deprecated) paths:
`PUT /v1/agents/{id}/call-controls`
scheduleJob pathParams, bodyParams
**Deprecated:** use `POST /v1/calls`, which returns the call itself.

Schedule a new outbound call job for a voice agent.

The job can be scheduled for a future time or triggered immediately with `trigger_immediately: true`.

**Scheduling:**

When scheduling for a future time, provide `scheduled_at` (ISO 8601) and optionally `schedule_window_minutes` (minimum 10 minutes).

**Environment:**

`environment` selects which promoted agent configuration and phone binding the call dispatches against (`live` or `test`). Defaults to `live`. Scheduling fails when the selected environment has no active promotion.

For compatibility this endpoint is also available using the following (deprecated) paths:
`POST /v1/agents/{id}/jobs`
listJobs queryParams
**Deprecated:** use `GET /v1/calls?state=scheduled,in_progress,ended,canceled,failed`, which also filters by `scheduled_after` / `scheduled_before`. `GET /v1/calls/{id}` returns a call's `payload`.

List all outbound jobs with pagination and filtering.

**Filtering:**

- Filter by agent ID, status, or scheduled time range.
- Multiple statuses can be comma-separated (e.g. `status=pending,queued`).

For compatibility this endpoint is also available using the following (deprecated) paths:
`GET /v1/agents/jobs`
cancelJob pathParams
**Deprecated:** use `POST /v1/calls/{id}/cancel`.

Cancel a pending or queued outbound job. Jobs that are already in progress or completed cannot be canceled.

For compatibility this endpoint is also available using the following (deprecated) paths:
`POST /v1/agents/jobs/{id}/cancel`
listApiKeys queryParams
List all API keys for your tenant.

By default, revoked keys are excluded. Set `include_revoked=true` to include them.

**Note:** The full API key value is never returned — only the key prefix is shown.
createApiKey bodyParams
Create a new API key for authenticating with the Avallon API.

**Important:** The full API key is only returned once in the response. Store it securely — it cannot be retrieved again.

Keys are scoped to an environment (`test` or `live`) and optionally accept an expiration date.
revokeApiKey pathParams
Revoke an existing API key. Once revoked, the key can no longer be used for authentication.

This operation is irreversible — revoked keys cannot be re-enabled. Create a new key if needed.

Returns 409 if the key is already revoked, 404 if the key does not exist.
updateApiKey pathParams, bodyParams
Update mutable fields of an API key. Currently only `role` can be updated.

The caller's role caps the maximum role they can grant — an `admin` cannot promote a key to `avallon_admin`.
createArtifact bodyParams
Upload a document as a base64-encoded artifact for processing.

**Limits:**

Maximum file size is approximately 6.75MB (base64 encoded). The API Gateway has a 10MB request limit.

**Processing options:**

- `file_password`: Password for encrypted PDFs
- `file_splitting_enabled`: Split multi-page PDFs into separate artifacts
listArtifacts queryParams
List all artifacts with pagination, filtering, and sorting.

**Case Filtering:**

Use `case_id` to return only artifacts linked to a case. This reads the case links
created via `POST /v1/cases/{caseId}/links`, and is independent of any
`metadata.case_id` an artifact may carry.

**Metadata Filtering:**

Use bracket notation to filter by metadata keys:
- `metadata[claim_id]=abc-123`
- `metadata[source]=email`

Multiple metadata filters can be combined (AND logic).
getArtifact pathParams
Retrieve a single artifact by ID, including its metadata, file URL, extracted content, and chunk-annotated content.

The response always includes `content_with_chunkids_md`: the parsed content annotated with `[chunk_id=...]` markers when the artifact has parsed chunks. Chunk IDs are stable citation anchors — cite them to attribute extracted values to their source location.
deleteArtifact pathParams
Soft-delete an artifact. Soft-deleted artifacts and their chunks no longer appear in any read.
getArtifactChunks queryParams
Resolve chunk ids to their document, page, bounding box and text, keyed by chunk id.

Chunk ids are the `[chunk_id=…]` markers in `content_with_chunkids_md` and the values
workers cite (e.g. `data-cite-chunk`). Ids that are unknown or whose document was deleted
are omitted.
getArtifactStatusSummary queryParams
Count the artifacts in a scope, broken down by processing status.

Define the scope with bracket notation (e.g. `scope[email_id]=abc-123`). At least one key is required; multiple keys are combined with AND logic.
getArtifactUploadUrl bodyParams
Returns a pre-signed PUT URL. Upload the file directly to `upload_url`, then pass `file_url` as `source_url` to POST /v1/upload-file.
updateArtifactMetadata pathParams, bodyParams
Merge-update metadata on an existing artifact. Provided keys are added or overwritten; other existing keys are preserved.
signUp bodyParams
Create a new user account with email and password. When the response includes a pending_authentication_token, the account requires email verification: complete it via sign-in (type: email_otp) with the emailed one-time password and this token.
signIn bodyParams
Authenticate using one of the supported methods:

- **Password:** Email and password credentials. If the password is correct but the email is not verified yet, responds 403 with `data.pending_authentication_token` and a fresh verification code is emailed; complete verification via **Email OTP**. Each password attempt invalidates earlier codes and tokens, so always use the most recent pair — repeating the password sign-in is also how a new code is requested.
- **OAuth:** Authorization code from OAuth flow (social or enterprise SSO). Pass the `provider` the flow was started with so the code is exchanged against the system that issued it. `verifier` is only required when the flow was started with a `code_challenge` — see **Get OAuth sign-in URL**.
- **Email OTP:** One-time password sent to email. Include the `pending_authentication_token` from sign-up, or from the 403 returned by an unverified user's password sign-in, when completing email verification.
requestPasswordReset bodyParams
Start password recovery for an email/password account. When the email belongs to an account, a reset link is emailed to it. The response is 200 regardless, and does not reveal whether an account exists.
resetPassword bodyParams
Complete password recovery with the token from the emailed reset link and set a new password. Completing the reset also verifies the account's email address. Afterwards, sign in with the new password.
refreshToken bodyParams
Exchange a valid access token and refresh token pair for new tokens.
refreshSession bodyParams
Rotate the refresh token and return a new access token for `tenant_id`. Call it with a different `tenant_id` to switch tenants.

Send the current access token as `Authorization: Bearer`. It may be expired: it only identifies the user, and the refresh token must belong to that same user. The identity provider decides whether the session is still valid. The user must be a member of `tenant_id` (404 otherwise).
getOAuthUrl queryParams
Generate an OAuth authorization URL for a given provider.

PKCE is optional. To use it, generate a random verifier, derive `code_challenge` from it, and pass both `code_challenge` and the matching `code_challenge_method` — then send the verifier to **Sign in** when exchanging the code. Omit both to run the flow without PKCE and exchange the code on its own.

For enterprise SSO, pass `provider=sso` together with the user's `email`; the email domain selects the SSO organization. Responds 404 when no SSO organization is configured for the domain.
getSessionToken bodyParams
Exchange an identity token for a tenant-scoped Core Plane access token.

- If `tenant_id` is provided, validates the user is a member of that tenant and returns 404 if they are not.
- If `tenant_id` is omitted, selects the tenant of the identity token's organization, falling back to the user's default (oldest) tenant.
- Call this endpoint again with a different `tenant_id` to switch tenants.
getPartnerJwks pathParams
Returns the JSON Web Key Set (RFC 7517) of a partner's public signing keys, used to verify partner-issued launch tokens.
partnerLaunch queryParams
Verifies a partner-signed launch token, mints an Avallon session, and redirects (302) the user into the app's handoff route, carrying the session token in the URL fragment (no cookie).
listCalls queryParams
List all calls, with pagination, filtering, and sorting.

**Filtering:**

Filter by agent, direction, status, time range, phone number, and more. Phone number filter supports both exact E.164 matches and partial search.

**Case Filtering:**

Use `case_id` to return only calls linked to a case. This reads the case links created
via `POST /v1/cases/{caseId}/links`, and is independent of any `metadata.case_id` a
call may carry.

**Metadata Filtering:**

Use bracket notation to filter by metadata keys:
- `metadata[claim_id]=abc-123`
- `metadata[source]=worker`

Multiple metadata filters can be combined (AND logic).
createCall bodyParams
Create an outbound call for a voice agent, either for later or right away.

**Scheduling:**

Provide `scheduled_at` (ISO 8601) to place the call later. It is dialed within 60 minutes of that time, or it expires. Omit `scheduled_at` to dial immediately.

**Lifecycle:**

The call starts as `scheduled` and moves to `in_progress` when it is dialed, then `ended`. A scheduled call can be canceled with `POST /v1/calls/{id}/cancel`.

**Environment:**

`environment` selects which promoted agent configuration and phone binding the call dispatches against (`live` or `test`). Defaults to `live`.
getCall pathParams, queryParams
Retrieve detailed information about a specific call.

Set `include_messages=true` to include the full message transcript.
cancelCall pathParams
Cancel an outbound call that is still `scheduled`. Canceling a call that is already canceled returns it unchanged; a call that is in progress or has ended cannot be canceled.
listenCall pathParams
Issue a short-lived token that joins a connected call as a hidden, listen-only participant. Only calls that are `in_progress` with a `connected_at` can be listened to.
getCallEvaluation pathParams
Retrieve the evaluation results for a specific call, including overall score, pass status, and per-metric breakdowns.
getCallAnalytics queryParams
Retrieve aggregated call analytics for a time range.

Includes summary statistics, call volume over time, evaluation scores, score and duration distributions, and per-metric performance.
getAnalyticsTimeSeries queryParams
Retrieve a tenant-scoped time series for a single metric.

Pick a `metric`, a date range, a bucket `granularity` (day/week/month), and optionally a `group_by` dimension and filters. The series is sparse (only buckets with data) over the half-open window `[from, to)`.

- `worker_runs`, `documents`, `pages`: processing volume, bucketed by creation time. `worker_runs` can be grouped by `worker_id` and filtered by `worker_id` and `environment`. Documents are shared across environments and cannot be filtered by them.
- `finished_worker_runs`, `avg_worker_run_duration_ms`, `avg_worker_run_artifacts`, `worker_run_cost_usd`, `avg_worker_run_cost_usd`: worker runs that ended as completed, failed or canceled, windowed and bucketed by completion time. Runs created by worker evals and benchmarks are excluded. These can be filtered by `worker_id` and `environment`.
  - `finished_worker_runs`: number of finished runs. Can be grouped by `outcome` (`goal_reached`, `goal_not_reached`, `failed`, `canceled`); a completed run without a reported outcome counts as `goal_reached`.
  - `avg_worker_run_duration_ms`: average time from start to completion, in milliseconds. Runs that never started are excluded.
  - `avg_worker_run_artifacts`: average number of input artifacts resolved from the run's scope. Runs without a recorded artifact resolution are excluded.
  - `worker_run_cost_usd`, `avg_worker_run_cost_usd`: total and average estimated LLM cost of the runs, in USD. Canceled runs, runs that never reported the cost of their extraction, and runs that used a model without published pricing are excluded. Requires the `analytics:view-cost` permission.

`data` is an array of per-metric results; today it always contains exactly one entry (the requested metric).
createCallFeedback pathParams, bodyParams
Create feedback for a specific message in a call. Requires user (JWT) authentication; the author is derived from the authenticated user's profile and API-key callers are rejected.
listCallFeedback pathParams, queryParams
List all feedback for a call. Paginated.
deleteCallFeedback pathParams
Delete a single feedback entry.
checkInboxAvailability queryParams
Check whether a given email username is available for creating a new inbox.

For compatibility this endpoint is also available using the following (deprecated) paths:
`GET /v1/email-agents/availability`
createInbox bodyParams
Create a new inbox linked to a processor (extractor or worker).

The inbox receives emails at the live address built from `username` and at a test address built from `test_username` (defaults to `<username>-test`), and automatically processes attachments using the specified processor.

Either `processor_id` or the deprecated `extractor_id` must be provided. When using `extractor_id`, the processor type defaults to `extractor`.

For compatibility this endpoint is also available using the following (deprecated) paths:
`POST /v1/email-agents`
listInboxes queryParams
List all inboxes with pagination and sorting.

For compatibility this endpoint is also available using the following (deprecated) paths:
`GET /v1/email-agents`
updateInbox pathParams, bodyParams
Update an inbox's configuration.

For compatibility this endpoint is also available using the following (deprecated) paths:
`PATCH /v1/email-agents/{id}`
deleteInbox pathParams
Delete an inbox. The inbox stops routing new mail; existing email history is preserved.
getEmail pathParams
Retrieve detailed information about a specific email, including full body content and the artifact IDs of its attachments.
deleteEmail pathParams
Soft-delete an email. The message is hidden from all reads (get, list, and thread aggregation) but its content is retained until a separate hard purge.
listEmails queryParams
List all emails with pagination and filtering.

**Filtering:**

Filter by email agent, sender, recipient, content search, or case.
listEmailThreads queryParams
List email threads with pagination. Each thread includes the subject of the first email, the sender of the first email, and the date of the last email.

**Filtering:**

Filter by inbox or environment. Without an `environment` filter, live and test threads are returned together.
replyToEmail pathParams, bodyParams
Reply to an existing email thread via Nylas. The reply is sent to the original sender by default.

Optionally include additional recipients via `to` (merged with the original sender), `cc`, and `bcc` arrays.

The `body` is HTML by default. Set `is_html` to `false` to provide plain text instead; it is converted to HTML before sending.
sendEmail bodyParams
Send a new email through an inbox. The provider-assigned thread id is returned and persisted on the stored email row, so subsequent inbound replies on that thread are linked to this send.

The `body` is HTML by default. Set `is_html` to `false` to provide plain text instead; it is converted to HTML before sending.
listExtractorJobs queryParams
List all extraction jobs with pagination, filtering, and sorting.

**Scope Filtering:**

Use bracket notation to filter by scope keys:
- `scope[call_id]=abc-123`
- `scope[email_id]=xyz-456`
- `scope[case_id]=case-789`

Multiple scope filters can be combined (AND logic).
createExtractorJob bodyParams
Create a new extraction job for the specified extractor.

Provide a non-empty `scope` object to define what the extractor should process (e.g. `{ "claim_id": "abc-123" }`).
getExtractorJob pathParams
Retrieve detailed information about a specific extractor job, including its status, timing, and result.
listWorkerRuns queryParams
List all worker runs with pagination, filtering, and sorting.

**Scope Filtering:**

Use bracket notation to filter by scope keys:
- `scope[call_id]=abc-123`
- `scope[email_id]=xyz-456`

Multiple scope filters can be combined (AND logic).
createWorkerRun bodyParams
Create a new worker run for the specified worker.

Provide a `scope` object with at least one key to define which artifacts the worker should process (e.g. `{ "claim_id": "abc-123" }`). The run stays `pending` until those artifacts are ready.

Pass `null` to run the worker without any artifacts — the run is queued immediately. An empty object, or a scope that matches no artifacts, is rejected.
getWorkerRun pathParams
Retrieve detailed information about a specific worker run, including its status, timing, and errors.
listWorkerRunMessages pathParams, queryParams
Retrieve messages emitted during a worker run, ordered by emission time. Supports cursor-style pagination via `after` and `count`.
cancelWorkerRun pathParams
Cancel a worker run. For in_progress runs, the underlying E2B sandbox is terminated when possible. Pending and queued runs are simply marked canceled. Runs that are starting are still being handed their context and cannot be interrupted, so they cannot be canceled until they reach in_progress. Runs in terminal states (completed, failed) cannot be canceled. Already-canceled runs are returned without side effects.
createWorkerEvalFromRun pathParams, bodyParams
Snapshot a worker run's trigger and resolved artifacts, and record its MCP tool responses from telemetry for replay. Each assertion is graded PASS/FAIL independently on every run.
runEval pathParams, bodyParams
Queue an explicit worker version against the eval's immutable artifact snapshot and mocked tools.
getEvalRun pathParams
Get the status and prompt-grader result for an eval run.
createEvalBenchmark pathParams, bodyParams
Queue runs_per_configuration trials of the eval against a candidate and a baseline worker version. When all trials finish, the benchmark reports mean ± stddev pass rate per configuration, the candidate-minus-baseline delta, and analyzer notes.
listEvalBenchmarks pathParams, queryParams
Benchmark history of an eval (newest first by default) without per-run details. Use GET /v1/benchmarks/{benchmarkId} for the runs.
getBenchmark pathParams
Get a benchmark with its per-run gradings and metrics, and once all runs finished, the per-configuration summary, delta, and analyzer notes.
listWorkerEvals pathParams, queryParams
List the evals recorded for a worker, newest first by default.
createWorkerBenchmarks pathParams, bodyParams
Create one benchmark per eval of the worker, each queueing runs_per_configuration trials against the candidate and the baseline version. Poll each benchmark with GET /v1/benchmarks/{benchmarkId}.
createExtractor bodyParams
Create a new extractor with a name and JSON schema defining the extraction output format.
listExtractors queryParams
List all extractors with pagination, filtering, and sorting.

Filter by name using the `name` query parameter.
getExtractor pathParams
Retrieve a single extractor by ID, including its schema and version.
updateExtractor pathParams, bodyParams
Update an extractor's schema. This creates a new version of the extractor.
deleteExtractor pathParams
Soft-delete an extractor and all of its versions. Soft-deleted extractors no longer appear in any read.
createWorker bodyParams
Create a new worker with a name and prompt.
listWorkers queryParams
List all workers with pagination, filtering, and sorting.

Filter by name using the `name` query parameter.
getWorker pathParams
Retrieve a single worker by ID.
deleteWorker pathParams
Delete a worker. The worker is hidden from all reads; its run history is preserved.
updateWorker pathParams, bodyParams
Partially update a worker. Omitted fields are left unchanged, and one request creates at most one new config version.
getWorkerPrompt pathParams, queryParams
Retrieve the prompt for a worker. Pass `version` to fetch the prompt snapshot at a specific config version; omit it to fetch the current prompt. Internal use only.
updateWorkerPrompt pathParams, bodyParams
Update the prompt for a worker. Internal use only.

**Deprecated:** Use `PATCH /v1/workers/{workerId}` instead.
listWorkerModels
List the LLM models that can be assigned to a worker via the `model` field on update.
updateWorkerMcps pathParams, bodyParams
Replace the set of MCP servers a worker connects to. Pass an empty array to disconnect.

**Deprecated:** Use `PATCH /v1/workers/{workerId}` instead.
getAccess pathParams
List who an entity is shared with. Admins see every entity through their role, whether or not they are listed. Requires `workers:write` on a worker, `case-types:write` on a case type and `voice-agents:write` on a voice agent.
changeAccess pathParams, bodyParams
Grant, change or revoke access for specific team members and workers. Everyone not named keeps their access. Each grantee must be a team member or a worker of the tenant. Requires `workers:write` on a worker, `case-types:write` on a case type and `voice-agents:write` on a voice agent. An editor who revokes their own grant loses access, and one who downgrades it can no longer change or share the entity, unless their role gives it.
getMyAccess pathParams
List what the caller may do with an entity, through their role or a grant on it, to decide which controls to show. An entity the caller cannot see is not found.
listWorkerVersions pathParams, queryParams
List all config versions for a worker, ordered by version (newest first by default).
getWorkerVersion pathParams
Retrieve a single worker config version, including the worker's identity, the config snapshot at that version, and version-specific metadata. The prompt is served separately by the worker prompt endpoint.
promoteWorkerVersion pathParams, bodyParams
Select a worker configuration version for the live or test environment. This freezes the config snapshot only — the E2B template, attached tools and attached secrets are not versioned and always resolve to their current values.
getWorkerVersionChangeSummary pathParams
Returns an AI-generated summary of what changed between this version and the previous one. Initial version returns an empty string since there is no predecessor.
createWorkerWebhook pathParams, bodyParams
Create a webhook that fires when events occur for the specified worker.
listWorkerWebhooks pathParams, queryParams
List all webhooks configured for a specific worker.
listCases queryParams
List all cases for the current tenant, paginated, filterable, and sortable.
createCase bodyParams
Create a case with a client-supplied id.

Case ids are tenant-scoped opaque keys (e.g. "clm-1", "STD-001") and are normalized server-side (trimmed and lowercased). Creating an id that already exists **in the same environment** returns 409 — the same id can exist once in `live` and once in `test`.
searchCases queryParams
Probe for candidate cases. Matches the query as a normalized substring of the case id, and fuzzily (trigram) against subject and reporter. Results are relevance-ranked — id matches first, then best fuzzy match, then most recently updated — so ordering is fixed and not configurable.
getCase pathParams, queryParams
Retrieve a case by id, with its current fields.

Related entities are not embedded — fetch them via the filtered list endpoints,
e.g. `GET /v1/worker-runs?case_id=…` and `GET /v1/artifacts?metadata[case_id]=…`.
deleteCase pathParams, queryParams
Delete a case.

The case and its links stop being visible; the linked entities themselves are untouched. The id becomes available again — a case created with the same id afterwards is a new, empty case, not the deleted one.

Unlike removing a case link, this is **not** idempotent: deleting a case that does not exist, or has already been deleted, returns 404.
updateCase pathParams, queryParams, bodyParams
Only the fields sent change.
getCaseCitations pathParams, queryParams
Resolve the chunks cited by a case's current field values, keyed by chunk id.

Each field in `GET /v1/cases/{id}` lists its citations as chunk ids; this endpoint
returns the document, page, bounding box and text for each. Chunks whose document
was deleted are omitted. A case with no cited fields returns an empty map.
addCaseLinks pathParams, queryParams, bodyParams
Link up to 100 entities (artifacts, calls, chats, emails, extracts, information requests, schedules, worker runs) to a case.

The case and every entity must already exist in the tenant — unknown ids return 404 and nothing is created. Calls, emails, information requests and worker runs must belong to the case's environment, and a schedule must run in it; artifacts, chats and extracts are not environment-scoped and can be linked to a case in either. A mismatch rejects the whole request. Adding is idempotent: re-linking an already linked entity returns the existing link with its original `created_at`.
listCaseLinks pathParams, queryParams
List the entities linked to a case, paginated and filterable by entity type.
removeCaseLink pathParams, queryParams
Remove the link between a case and an entity. Removal is idempotent: removing a link that does not exist succeeds without effect.
createChatAgent bodyParams
Create a new chat agent with a name and prompt.
listChatAgents queryParams
List the tenant's chat agents with pagination and sorting. Prompts are not returned.
getChatAgent pathParams
Retrieve a single chat agent by ID, including the IDs of its attached code tools. The prompt is not returned.
updateChatAgent pathParams, bodyParams
Patch a chat agent's attached code tool set and evaluator. The response does not include the resulting tool list; the agent's `updated_at` is not bumped when only the tool set changes.

Set `evaluator_slug` to score the agent's chats with that evaluator when they end. It applies to every chat that ends afterwards.
deleteChatAgent pathParams
Soft-delete a chat agent. The agent no longer appears in any read and stops accepting new chats; existing chats and messages are preserved.
getChatAgentPrompt pathParams
Retrieve the prompt and greeting for a chat agent. Internal use only.
updateChatAgentPrompt pathParams, bodyParams
Update the prompt and/or greeting for a chat agent. Internal use only.
createChat bodyParams
Create a new chat against an existing chat agent. The chat agent's tenant becomes the owner of the chat. If the agent requires authentication (`identity_provider_id` on the public chat agent), send the identity token from `POST /public/v1/identity/token` in the `X-External-Identity-Token` header; without a valid token the request fails with 401.
getChat pathParams
Return a chat and its messages in chronological order. Used to re-render a transcript after a reload. Only `user` and `assistant` messages are included; tool calls and tool results are omitted. Each message's `content` is the concatenation of its text blocks separated by newlines.

For compatibility this endpoint is also available using the following (deprecated) paths:
`GET /public/v1/chat`
getPublicChatAgent pathParams
Return the public-facing fields of a chat agent for use in the chat UI. Sensitive fields such as the agent's prompt and tenant ID are not exposed.
postChatMessage pathParams, bodyParams
Post a user message to an existing chat and return the assistant reply. Both messages are persisted.

For compatibility this endpoint is also available using the following (deprecated) paths:
`POST /public/v1/chat/messages`
endChat
End a chat. Further messages are rejected with `409`. Ending an ended chat also returns `204`.
exchangeExternalIdentityToken bodyParams
Redeem a single-use interclient token from a partner identity provider for a short-lived Avallon identity token. A malformed request body returns 400. A rejected exchange returns 401; the interclient token may already be spent, so request a fresh one before retrying.
listChats queryParams
List the tenant's chats with pagination, sorting, and optional filtering by case. Returns chat metadata, message counts and evaluation outcome; messages are not included.
listChatMessages pathParams, queryParams
List messages for a chat owned by the tenant, in chronological order by default.
getChatEvaluation pathParams
Retrieve the evaluation results for a chat, including overall score, pass status, and per-metric breakdowns.
getPublicFormSubmission pathParams, queryParams
Return the title and JSON Schema of the form version a submission link is pinned to, for the form renderer. The id and the password set on create are the credential; an unknown id is 404 and a wrong password, or a link created without one, is 401. Once the request has been submitted the request fails with 409 and `{ title }`.
submitFormSubmission pathParams, bodyParams
Store the answers for a submission link and hand them to the form's processor. Each link can be submitted once: a second submit fails with 409. The id and the password set on create are the credential; an unknown id is 404 and a wrong password, or a link created without one, is 401.
getPublicFormSubmissionTheme pathParams
Return the form title and the name and branding of the tenant a submission link belongs to, so the form renderer can style the page before the password is entered. Needs no password; an unknown id is 404.
createForm bodyParams
Create a form. Its schema becomes version 1, used by both live and test. The schema is validated against the renderer subset; violations are reported as 400 with the offending field path.
listForms queryParams
List the forms shared with the caller, each at its latest version.
getForm pathParams
Retrieve a form at its latest version.
updateForm pathParams, bodyParams
Partially update a form. Omitted fields are left unchanged, and one request creates at most one new version — none when nothing changed. Live and test keep pointing at their versions until promoted.
listFormVersions pathParams, queryParams
List a form's versions without their schemas, newest first by default. Fetch a version's schema with the form version endpoint.
getFormVersion pathParams
Retrieve a form at the given version, including that version's schema.
promoteFormVersion pathParams, bodyParams
Point the live or test environment at an existing form version. New submission links for that environment are pinned to it; open links keep their version.
createFormSubmission bodyParams
Open a submission link for a form, pinned to the version the given environment points at. Send the returned url and the password to the person filling the form.
listFormSubmissions queryParams
List the submission links of the forms shared with the caller, with the answers of those already submitted.
getFormSubmission pathParams
Retrieve a submission link and, once it was submitted, the answers. Used by the form's worker and the studio; respondents use the public endpoint instead.
getExtract pathParams
Retrieve a single extract by ID, including the extraction result data.
getExtractCitations pathParams
Retrieve the citation tree with resolved chunk data for a given extract. Returns 404 if the extract does not exist or has no citations (the citations sub-resource is not available).
listExtracts queryParams
List all extracts with pagination, filtering, and sorting.

**Scope Filtering:**

Use bracket notation to filter by scope keys:
- `scope[claim_id]=abc-123`

Multiple scope filters can be combined (AND logic).
getProfile
Retrieve the current user's profile information. This endpoint requires a user session and cannot be used with API Keys.
listTeamMembers
List all members in the current tenant.
inviteTeamMember bodyParams
Invite an email address to join the current tenant. The invitee receives an email and joins at the `member` role once they accept; promote them afterwards with `PATCH /v1/team/members/{id}`.

Requires the `teams:write` permission.

Returns `409` if the address already has a pending invitation or is already a member.
listTeamInvitations
List the invitations for the current tenant that are still awaiting acceptance. Accepted, revoked and expired invitations are not returned.

Invitees are not team members yet, so they do not appear in `GET /v1/team/members`. A member-management screen shows both by merging the two lists.

Requires the `teams:write` permission.
revokeTeamInvitation pathParams
Cancel a pending invitation. The link that was emailed to the invitee stops working, and the address becomes free to invite again.

Idempotent: an invitation that was already accepted, revoked or expired also returns `204`.

Requires the `teams:write` permission.

Returns `404` for an unknown invitation or one that belongs to another tenant.
resendTeamInvitation pathParams
Email the invitation to its recipient again, for an invite that never arrived. The invitation keeps its `id` and its original expiry — this re-sends the existing link rather than issuing a new one.

To give someone a genuinely fresh 5 days, revoke the invitation and send a new one instead.

Requires the `teams:write` permission.

Returns `409` if the invitation is no longer pending — it was accepted, revoked or expired underneath the caller — and `404` for an unknown invitation or one that belongs to another tenant.
updateMemberRole pathParams, bodyParams
Change the role of a team member.
removeMember pathParams
Soft-delete a team member, revoking access on the next token refresh.
getTenantSettings
Retrieve the settings of the current tenant.
updateTenantSettings bodyParams
Update settings of the current tenant. Requires the `tenants:write` permission, which only Avallon admins hold.
createSecret bodyParams
Create an organization-level secret. The value is encrypted at rest and **write-only** — no endpoint ever returns it.

Requires the `secrets:write` permission. Env keys used by the worker runtime itself (e.g. `OPENAI_API_KEY`, `AVALLON_*`, `AWS_*`) are reserved and rejected.
listSecrets queryParams
List the organization's secrets with pagination and sorting. Values are never included. Requires the `secrets:read` permission.
getSecret pathParams
Get a secret's metadata. The value is never included. Requires the `secrets:read` permission.
updateSecret pathParams, bodyParams
Update a secret's display name or description, or rotate its live or test value. Rotation overwrites the stored value — previous values are not retrievable.

The `env_key` is immutable: consumers of the secret receive the value under that key, so changing it would silently break them. Create a new secret instead.

Requires the `secrets:write` permission.
deleteSecret pathParams
Delete a secret and detach it from all entities. Requires the `secrets:write` permission.
createEvaluator bodyParams
Create an evaluator: a named set of metrics that an LLM judge scores every call or chat of the agents it is attached to.

Evaluators can't be changed after creation, so past scores keep their meaning. To change the criteria, create a new evaluator (e.g. `support-quality-v2`) and attach that instead. A slug that already exists returns 409.

Attach it to a voice agent with `PATCH /v1/voice-agents/{voiceAgentId}` (`evaluator_slug`), then promote that version, or to a chat agent with `PATCH /v1/chat-agents/{chatAgentId}` (`evaluator_slug`). Requires the `evaluators:write` permission.
listEvaluators
List all evaluators of the tenant with their metrics.
getEvaluator pathParams
Retrieve an evaluator and the metrics its judge scores.
createSchedule bodyParams
Create a schedule that performs an action, either repeatedly or once.

`action` describes **what** happens: `trigger_worker` starts a worker run with the given scope, and `execute_tool` executes a tool without parameters. `rule` describes **when**: `recurring` carries one or more `patterns`, and fires whenever any of them comes round. A pattern fires every hour at `minute`, narrowed by `hours` and `weekdays` — so `hours: [9]` is daily at 09:00, `hours: [12], weekdays: [1]` is Mondays at noon, and `hours: [9, 13, 17], weekdays: [1, 2, 3, 4, 5]` is three times a day on weekdays. Give a second pattern to mix cadences: hourly on weekdays plus 09:00, 11:00 and 18:00 at weekends is two patterns on one schedule. Patterns overlap harmlessly — two that land on the same minute fire once. Every set is read in the schedule's own time zone — an IANA identifier such as `America/Chicago`, resolved across DST transitions. `once` fires a single time at a UTC instant.

`environments` says **which version** of the worker or tool each occurrence runs: `live` resolves it at its live version, `test` at its test version. Give both to soak a candidate version on the same cadence as production — every occurrence then fires twice, once per environment. Omit it to fire in `live` only.

The response's `next_run_at` is the first occurrence strictly after the moment of creation; a rule that resolves to no such occurrence is rejected. Create with `enabled: false` to stage a recurring schedule without letting it fire. A one-off is always enabled: delete it to cancel it, and its `next_run_at` becomes null once it has fired.

Requires the `schedules:write` permission, and `tools:execute` for an `execute_tool` action.
listSchedules queryParams
List the organization's schedules. Requires the `schedules:read` permission.
getSchedule pathParams
Retrieve a single schedule. Requires the `schedules:read` permission.
updateSchedule pathParams, bodyParams
Update a schedule's rule, environments, name, or enabled state.

Changing the rule moves `next_run_at` to the first occurrence after the update; a rule that resolves to no such occurrence is rejected. One-off schedules are always enabled — delete one to cancel it before it fires.

`environments` replaces the whole list rather than adding to it. It takes effect from the next occurrence: one already due keeps the environments it was created with.
deleteSchedule pathParams
Delete a schedule so it stops firing.
listDispatches queryParams
List the organization's dispatches, newest occurrence first.

A dispatch is one occurrence of a trigger in one environment, carried out on its own: a schedule produces one per environment it runs in every time it fires, an inbound email produces one for the environment its inbox received it in, a resolved information request produces one that restarts the worker that asked, an ended call produces one that runs its voice agent's post-call processor once its transcript is ready, and creating a worker run through the API produces a `manual` one, carried out before the request returns. Listing them shows what a trigger has actually done, including occurrences that never produced anything.

- `pending` — not carried out yet: waiting to be picked up, in flight, or waiting on a retry after an earlier pickup did not complete.
- `dispatched` — carried out.
- `abandoned` — gave up, because the occurrence passed its `expires_at` deadline before it could be carried out, or, for a `manual` dispatch, because the request that created it failed.

`attempts` counts how many times the occurrence has been picked up, not how many times it failed: a dispatch carried out on the first pickup ends at `attempts: 1`, and a higher count means earlier pickups did not complete.

`target` is what the occurrence produced: a worker run, an extractor job (for an email whose inbox extracts or a call whose voice agent extracts), or a tool call (for an `execute_tool` schedule). It is null until the dispatch is carried out, and for a tool call until the tool has run.

Narrow to one trigger with `trigger_type` and `trigger_id` — the schedule's ID for a schedule, the email's ID for an email, the request's ID for an information request, the call's ID for a call. `trigger_id` requires `trigger_type`.
listNotifications queryParams
List the organization's notifications, newest first.

A notification is a durable record that something happened and that somebody should know about it. Notifications are not yet addressed to individual people, so everyone in the organization sees the same feed.

`title` and `body` are the display text — render them as-is rather than deriving your own wording from `type`, so every surface says the same thing. `data` carries the identifiers the notification points at, so a client can link to the thing it is about; which keys are present depends on `type`.

`worker_issue_reported` means a worker hit a problem during a run and reported it; its `body` is the worker's own message and its `data` is the reporting `worker_id` and `run_id`. `information_request_created` means a worker paused to ask a person a question; its `data` is the `request_id` to answer through `POST /v1/information-requests/{informationRequestId}/resolve`.
getNotification pathParams
Retrieve a single notification by its ID, the id a `notification.created` webhook delivery carries.

`title` and `body` are the display text — render them as-is rather than deriving your own wording from `type`, so every surface says the same thing. `data` carries the identifiers the notification points at, so a client can link to the thing it is about; which keys are present depends on `type`.

`worker_issue_reported` means a worker hit a problem during a run and reported it; its `body` is the worker's own message and its `data` is the reporting `worker_id` and `run_id`. `information_request_created` means a worker paused to ask a person a question; its `data` is the `request_id` to answer through `POST /v1/information-requests/{informationRequestId}/resolve`.
createInformationRequest bodyParams
Ask a person for a specific fact you cannot establish yourself.

Use this when one missing piece of information is blocking you and only a human can supply it — which claim a document belongs to, which of two policies applies, what an ambiguous reference means. Ask one answerable question per request, and write it so it stands on its own: whoever resolves it sees your title and your question and nothing else.

Immediately link the request to its case with `addCaseLinks` and `entity_type: "information_request"`. That is how it reaches the person working that case; unlinked, it sits in a queue nobody is looking at. Leave it unlinked only when you genuinely cannot tell which case it belongs to.

**This does not block or suspend you.** Nobody resolves it while this run is happening. Carry on with what you can still do, or stop, and say in your final report that you asked.

Do not use it for progress updates, or for anything you can work out yourself.
listInformationRequests queryParams
List the organization's information requests, newest first.

An information request is a specific question a worker could not settle on its own and needs a person to resolve. It stays open until somebody resolves it; there is no separate status field.
getInformationRequest pathParams
Retrieve a single information request by its ID.
deleteInformationRequest pathParams
Delete an information request, open or resolved..
resolveInformationRequest pathParams, bodyParams
Resolve an open information request by providing an answer/resolution.
createCaseType bodyParams
Create a case type. Its fields become version 1, used by both live and test.
listCaseTypes queryParams
List the organization's case types, without their fields.
getCaseType pathParams
Retrieve a case type with its fields at the newest version. Inside a worker run, returns the version the run's environment points at.
updateCaseType pathParams, bodyParams
Update a case type. Its fields are changed through `/v1/case-types/{caseTypeSlug}/fields`.
listCaseTypeVersions pathParams, queryParams
List a case type's versions, newest first by default.
getCaseTypeVersion pathParams
Retrieve a case type with its fields at a specific version.
promoteCaseTypeVersion pathParams, bodyParams
Point the live or test environment at a version. Any existing version can be promoted, including an older one.
createCaseTypeField pathParams, bodyParams
Add a field to a case type. Creates a new version; promote it to use the field.
updateCaseTypeField pathParams, bodyParams
Change a field's label, description, nullability, enum or sensitivity. Creates a new version when something changes; promote it to use the change. A field's key, type, list-ness and parent are permanent.
deleteCaseTypeField pathParams
Remove a field from a case type. Creates a new version; once it is promoted, the field can't be written. Deleting an object deletes its fields with it. Cases keep their values, and the key can't be reused.
createTool bodyParams
Create a new code tool that can be attached to agents. Provide the tool's name, code, and description.
listTools queryParams
List all code tools with sorting.

For compatibility this endpoint is also available using the following (deprecated) paths:
`GET /worker/v1/tools`
deleteTool pathParams
Delete a tool. Fails with a conflict when the tool is still attached to a voice agent, worker, or chat agent; detach it first.
getTool pathParams
Retrieve a single tool by ID at its newest config version, including its code, schema, description and attached secrets, plus the versions each environment runs.

For compatibility this endpoint is also available using the following (deprecated) paths:
`GET /worker/v1/tools/{id}`
updateTool pathParams, bodyParams
Update a code tool's name, description, code, or attached secrets. All fields are optional — only provided fields are updated.

Changing the code, description or attached secrets creates a new config version; runs keep executing the versions promoted to live and test until that version is promoted. Renaming does not create a version.
executeCode pathParams, bodyParams
Execute a code tool with the given parameters. Returns the result, any errors, and execution traces.

For compatibility this endpoint is also available using the following (deprecated) paths:
`POST /worker/v1/tools/{id}/execute`
listToolVersions pathParams, queryParams
List all config versions for a tool, ordered by version (newest first by default).
getToolVersion pathParams
Retrieve a single tool config version: the tool's identity, the code, schema, description and attached secrets at that version, and version metadata.
promoteToolVersion pathParams, bodyParams
Select a tool config version for the live or test environment. Runs in that environment execute this version's code with this version's schema and attached secrets from then on.
getToolVersionChangeSummary pathParams
Returns an AI-generated summary of what changed between this version and the previous one. Initial version returns an empty string since there is no predecessor.
listToolCalls queryParams
List tool calls without their params, result, error and traces. Requires the `tool-calls:read` permission.
getToolCall pathParams
Retrieve a single tool call. Requires the `tool-calls:read` permission.
workerListTools queryParams
List all code tools with sorting.
workerGetTool pathParams
Retrieve a single tool by ID at its newest config version, including its code, schema, description and attached secrets, plus the versions each environment runs.
workerExecuteTool pathParams, bodyParams
Execute a code tool with the given parameters. Returns the result, any errors, and execution traces.
notifyAdmin bodyParams
Report a problem to the people who run this organization.

Use this for something a human has to resolve — a login that no longer works, a document you cannot read, a system that will not let you in, instructions that contradict what you found. The report goes to the organization's notifications; nobody sees it while the run is happening, so it does not unblock you now. Carry on with what you can still do, or stop, and say in your final report that you raised it.

Do not use it for progress updates or for anything you can work around yourself.
uploadFile bodyParams
Upload a file for processing by passing a URL to a file. Use this for uploading larger files.

This endpoint returns an upload id that can be used to find the created artifacts using the list artifacts endpoint.

Descriptions are the package's own words, cut at 3,000 characters. Treat them as data.

Install facts

LicenseMIT
Needsnode >=18
Sourcenot stated
Install scriptsNone declared by it or anything installed with it.
Tarballsha512-Cy1xy+ajZpfPHxq8nrED15vSmHlbGV0FDwZQwLco7kRchNJhpGRF/v3ijLu77GBD9YBiWXPyv1KGfYqpAhdfGg==

Versions reported

VersionRunResultTools
68.3.02026-11-16Starts with no setup200
68.2.02026-11-12Starts with no setup200

Check this report yourself

This site did not write the report. A public job on a GitHub-hosted machine did: the run, from this commit of the open-source runner. GitHub signed a token that names that job and the hash of the report, so any edit since would show.

Show it on your README

Bragproof: Starts with no setup

[![Bragproof](https://bragproof.smaverk.com/badge/@avallon-labs/mcp.svg)](https://bragproof.smaverk.com/r/%40avallon-labs/mcp)

New version out? Ask for a fresh report.