Echo MCP tool reference

Get a ready recording's five-minute download URL and timing. Videos have no audio.

List a meeting's recordings.

Read meetings by ID: title, time, participants, notes, documents, URLs and recording availability. For what people said, use read_transcripts.

Find or list meetings by words, people, domains, tags or dates; use search_transcripts for what was said.

Prepend or append Markdown to a meeting's My notes, or replace it. Acts as you and requires editing access. For replace, pass expected_revision from read_meetings.note_revision. Returns the saved note_revision after cloud persistence; open collaborative editors merge the edit live. Retry uncertain results with identical arguments and the same operation_id.

Read what was said in meetings by ID: a timestamped, speaker-attributed script per recording. Use it for quotes and exact wording; follow next_offset for more.

Search what was said in meeting transcripts.

Apply one pending edit proposal, changing meeting content; ask the human before calling it.

Discard a pending edit proposal permanently; meeting content is unchanged.

List meeting edit proposals by meeting or status, with the expected_version apply and discard need.

Propose exact-text summary or transcript fixes and dictionary terms for one meeting: a durable, human-reviewable pending edit proposal that never changes meeting content; only apply_edit_proposal does.

Propose a full replacement Markdown summary for one meeting: a durable, human-reviewable pending edit proposal that never changes meeting content; only apply_edit_proposal does.

Create an artifact definition: a named brief Echo generates per meeting, after it or live.

Soft-delete one artifact definition; meeting artifacts it already generated are kept.

List the user's active artifact definitions, without instruction bodies.

Read one artifact definition in full with the version update and delete need.

Update one artifact definition by exact ID; omitted fields are kept.

Send text feedback to Echo only when the user explicitly asks; it does not attach meeting content. Reuse operation_id on retries.

Read a workspace's default sharing for new agents.

Set a workspace's default sharing for new agents; existing agents are unchanged.

Save a person's lasting preference about an agent; changes nothing now. Then say: Saved your feedback for improvement.

List the feedback people gave an agent, newest first, a page at a time: id, time, body (cut at 4,000 characters, truncated marks it), rating, classification, source, run_id and the proposal it answered. Pass next_cursor as cursor for the next page. Record new feedback with managed_agent_feedback.

Read an agent's purpose, trigger, status, version (expected_version) and open proposal; include=details adds setup, include=specification the native version. A slack_watch trigger row carries its live watch status (pending batch, sources, recovery) and the platform batch window.

List accessible agents and Slack routers as a paged index: name, aliases, status, purpose, trigger, last run and open proposal. Read setup with managed_agent_get.

Read one proposal: title and summary, report, decision and reason, base and head commits, test verdicts, the feedback it addresses, changed files with line counts, and their unified diff. The diff holds up to 48 KB; diff_omitted names files left out, and path reads one file's diff. Publish or reject it with its id and content_hash.

List an agent's proposals, newest first, a page at a time: id, title, state (open, publishing, failed, published, rejected or abandoned), origin (reviewer, agent, learning or editor), who decided and when, a rejection's reason, the version it published, content_hash and review_url. Pass next_cursor as cursor for the next page. Read one with managed_agent_proposal_get.

Draft instruction, skill or routing changes for review, based on managed_agent_get include=implementation; nothing changes until a person approves via managed_agent_publish.

Apply an open proposal only after a person who saw its diff (managed_agent_proposal_get) explicitly approves; enthusiasm or a judge report is not approval. Pass the proposal ID and the content_hash they saw. Returns the proposal's row.

Reject an open proposal when a person explicitly asks. Pass the proposal ID, the content_hash they saw and, when they said why, reason in their words; later proposals read it. Returns the proposal's row.

Close an idle run and archive its Anthropic session. Set stop_running=true to stop active work too.

Read one exact run. Defaults to a small status summary without provider reads. Request result={turn_index,request_id} for the exact answer and reports; turn_index=0 is the initial turn. Optional confirmed_posts, events, access and configuration require editor access and report unavailable/not_authorized separately. Each paged section owns its cursor. Events are credential-redacted; incomplete history cannot prove no calls occurred, and current access does not prove historical access. Never executes tools or changes the agent.

List an agent's recent runs, statuses, failure reasons and delivery receipts. Use managed_agent_run_get for one exact run and its optional result, confirmed posts, events or connection access.

Deprecated compatibility tool; use managed_agent_run_get with confirmed_posts. Reads confirmed explicit Slack posts (editors only). limit 1–5; before=next_cursor. Excludes host answers and ephemeral replies.

Deprecated compatibility tool; use managed_agent_run_start. Starts a chat, transcript or latest-call run. Reuse request_id only to retry the same request.

Deprecated compatibility tool; use managed_agent_run_get with result={turn_index,request_id}. Reads the public answer and reports for one exact run turn. turn_index=0 is the initial turn; follow-ups require their send-message request_id. pending is working; completed includes idle awaiting_reply turns.

Queue a follow-up on a hosted run, including while it is working. Messages run in order and replies stay in Echo. Reuse request_id only to retry; after poll_after_ms read managed_agent_run_get with the returned run_id and result={turn_index,request_id}. Echo Desktop delivers the reply automatically.

Start a direct conversation with source=chat and text, or run a hosted agent on a pasted transcript. Customer-owned agents run only from their own triggers. Reuse request_id only to retry; read managed_agent_run_get with the admitted run.id and result={turn_index:0} for its result. Echo Desktop delivers the reply automatically.

Deprecated compatibility tool; use managed_agent_run_list. Read an agent's recent runs and failure reasons; include_result=true adds output.

List all schedules that start new sessions for this agent, including existing recurring triggers and routines. Read IDs and revisions here before schedule_save. For continuing the current session later, use schedule_resumption instead.

Create or patch a schedule that starts a new session, with an optional prompt. Use cron for recurring work or at for a one-time run, plus timezone. New schedules default inactive; set enabled=true to activate. Missed recurring occurrences combine into one catch-up run; one-time work remains pending. Set ends_at only for an explicit deadline. Read schedule_list first; use a new schedule_id and revision 0 to create, the returned revision to edit. Null clears prompt or ends_at; other omissions preserve settings. Saving never runs work immediately. To continue this session later, use schedule_resumption instead.

List or read an agent's saved script candidates; never installs, runs or approves code.

Read an agent's audience, role and grants. View access exposes retained run inputs and outputs.

Replace an agent's audience and View/Edit grants, which cover retained history and connections; keep wanted users from managed_agent_sharing_get.

Change an agent's settings, triggers or active state; omitted fields stay. triggers replaces the set; trigger_changes patches triggers by kind (upsert merges enabled/params over the current row, remove deletes it) and cannot be combined with triggers. expected_version comes from managed_agent_get; a connection change sends only agent, expected_version, request_id and connection. Google: connection.kind=google with broker_policy; detach with connection_id null and no policy. Reuse an environment connection in the same workspace: connection.kind=environment, connection_id from managed_agent_connection_list, expected_revision, no server name or policy. The attachment has its own revocable connection; rotation affects every holder of the shared value. Prompt or skill drafts: managed_agent_propose. Builders may also set environment and external_target, and provision with expected_version=0.

Get a single-use link to connect or reconnect Echo's Slack app for a workspace. Never changes agents or disconnects Slack.

Save a fact the user explicitly stated, in first person; never infer one. If it is already saved, use update_memory.

Forget one saved memory, only when the user asks.

List facts the user saved about themselves and their contacts.

Rewrite a saved memory in place, e.g. to correct it.

Create a scoped Echo Pro API key for meeting reads or workspace imports; it is shown once.

List the caller's API keys and revisions, without secrets.

Permanently revoke an API key, effective next request; deploy any replacement first.

Create a blank meeting note only when the user asks; it does not record or invite. Reuse operation_id on retries.

Get the Echo meeting for an upcoming event on your connected Google or Outlook calendar, creating its note with the event title, time and participants if needed. Supply provider event_id or ical_uid plus the exact starts_at. Returns session_id for My notes tools. Does not schedule, invite or record. Repeated calls reuse the meeting.

Import an external meeting into an authorized workspace. Supply text, or enroll audio metadata and PUT the raw file to /v1/meetings/{session_id}/audio with a meetings:import API key. Echo can transcribe and diarize uploaded audio. Reuse source identity on retries; existing text is never overwritten or resurrected. No automatic workflow runs.

Move one meeting to another workspace; the old workspace loses access, named shares and links stay. Pass share_effect from list_workspaces; confirm each meeting with the user or use propose_only.

Create a manual Echo workflow (not a managed agent) only when asked; it never runs.

List account settings with values, sources and revision. Null timezone or week_start and theme=system follow the device.

Change one account setting only when the user asks; change privacy or data-sharing consent only on an explicit request for that setting.

List account-overridable experiments with source, override and effective states, and revision.

Turn one experiment on or off for this account, or clear its override (restoring the source default, not off), only when the user asks.

List workspaces with share_effect; resolve names here, never guess, right before move_meeting.

Echo's onboarding and product guide, for 'Onboard me to Echo', getting started or how-to help.

List an endpoint's recent deliveries.

Resend a delivery with its original webhook-id; de-duplicating receivers ignore it.

Register an https webhook for every meeting the caller can read in a workspace (Echo Pro); its secret is shown once.

Delete a webhook endpoint and its pending deliveries.

List the caller's webhook endpoints and revisions, without secrets.

Issue a new signing secret, shown once; the old one stays valid for 24 hours.

Send a fake meeting.transcript_ready event to an active endpoint.

Edit an endpoint's url, description or event types, or re-enable it.