oto.cx
grain

grain

grain.

A meeting recording — list, fetch, rename, tag, or share one.

ajouter à otoauth : —prefix : grain_*

setup & usage

⚠ grain api key

create an API key in Grain (app → Settings → Integrations → API — see the integration doc), then paste it into oto.

  • byo-only: no shared oto key
  • Personal Access Token (per user) or Workspace Access Token (admin, access to all workspace data) — both work the same way here, Grain applies the scope on its side
    • ▸ meetings, transcripts, sharing, webhooks

    grain gives access to meeting recordings, transcripts and organization data, through 5 tools:

  • "what are MY latest meetings?" → grain_recording(op="list", filter={"attendance": "hosted"}) — see the scope warning below, attendance is what restricts to "mine"
  • "give me the summary and highlights of this meeting" → grain_recording(op="get", recording_id="...", include={"ai_summary": true, "highlights": true})
  • "full transcript as plain text / subtitles" → grain_transcript(recording_id="...", format="txt"|"vtt"|"srt")
  • "rename / tag / share this meeting" → grain_recording(op="update"|"tag"|"share_user"|"share_team", recording_id="...", ...)
  • "download this meeting's file" → grain_recording_file(op="download", recording_id="...")
  • "notify my external system on every new meeting" → grain_hook(op="create", hook_url="https://...", hook_type="recording_added")
  • "also notify me of new highlights, with the transcript" → grain_hook(op="create", hook_url="https://...", hook_type="highlight_added", include={"transcript": true})
  • "which teams / users / meeting types exist?" → grain_org(op="users"|"teams"|"meeting_types")
  • "this team's Customer Support meetings that talk about onboarding" → grain_recording(op="list", filter={"team": "...", "meeting_type": "...", "title_search": "onboarding"}) — the three filters combine, confirmed live (ids via grain_org)
    • · ⚠️ a Personal Access Token ALSO sees other people's meetings

    Verified live on 2026-08-20 on a real workspace: grain_recording(op="list") without an attendance filter returns the share_state="public" recordings of the whole organization, not only those of the token holder — observed concretely: meetings recorded by colleagues, where the PAT holder was not even a participant, came up first (chronological sort). This is the behavior documented by Grain ("Personal notes" + "Public notes: workspace-visible notes"), not a bug — but it is surprising if you expect a "my meetings only" scope.

    To scope to "my meetings": pass filter={"attendance": "hosted"} (meetings hosted by the token holder) or "attended" (meetings they took part in) — confirmed live, these two values filter correctly. Without this filter, an agent that lists carelessly can surface a colleague's customer call in an answer.

    list_users also returns the full workspace directory (all members, not only the token holder) — normal, it is a directory, not meeting content.

    · verification & fixed bug
  • tested live on 2026-08-20 with a real Personal Access Token (a customer workspace): 20 of the client's 21 methods worked first time — list/get recordings, the 4 transcript formats, tag/untag, share/unshare user, update (rename), download (real file, 21 MB), create_upload_url (real pre-signed S3 URL), and the full webhook cycle (create against a real reachable URL, list, delete)
  • one real bug was found and fixed: share_with_team expected team_id in the JSON body (like share_with_user), not in the URL path as the doc initially suggested — the documented form really returns 404. Fixed and locked in again by a test
  • no OpenAPI spec is accessible for this API (openapi.json/docs.json/mint.json/llms.txt all return 403 on developers.grain.com, a constant WAF block) — the initial build came from reading doc pages, now largely confirmed by the live test above
  • grain_recording(op="get") uses a POST on Grain's side (not GET) — confirmed by both the doc and the live test
  • mutations (tag/untag, update, share_*/unshare_*, webhook deletion) return {"success": true}, not the updated object — confirmed live
  • highlights are NOT a separate tool — they appear in grain_recording via include={"highlights": true}
  • grain_hook covers Grain's whole webhook surface — list/create/delete is all that exists on Grain's side (no PATCH/update, confirmed on the doc page, this is not partial coverage)
  • grain_hook(op="create"): Grain tests the reachability of hook_url at creation — the URL must answer 2xx immediately or the call fails (confirmed live with a real test URL); for hook_type="highlight_added"|"highlight_updated", include accepts {"transcript": bool, "speakers": bool} (confirmed on the doc); story_* events are the ONLY way to observe Grain Stories — no REST endpoint lists/fetches them directly, the webhook payload IS the data
  • include={"hubspot": true} on grain_recording returns {"hubspot_company_ids": [...], "hubspot_deal_ids": [...]}; include={"participants": true} also carries hs_contact_id per participant (null if not linked) — confirmed live
  • file upload (raw bytes) is deliberately not exposed as an MCP tool — grain_recording_file(op="create_upload_url") gives the URL, sending the bytes happens outside the agent (the oto-core client carries upload_recording_file for scripted use; verified: it is a pre-signed S3 URL on a DIFFERENT host than api.grain.com, the Grain Bearer is never sent to it)
  • upstream error format is not documented by Grain, and no upstream error was deliberately triggered during the live test — the error messages here stay generic (HTTP code) until a real error case has shown the actual shape
  • Grain rate limit: 300 requests/minute, all routes combined
    • ⚙ multiple comptes

    this connector accepts multiple comptes: each stored credential becomes a named account (one name per compte), at your level, your team's or your org's.

  • the first one doesn't need a name; from the second on, each one carries its own (the first then takes the name "principal", renamable)
  • without specification, the agent takes the only stored compte, otherwise the one marked as default; otherwise the call is refused, naming the available comptes — never a random choice
  • target a compte for a call: _account="<name>" on the tool; list them: oto_identity(op='list', connector='grain') (scope='org' or scope='group' for the org's or team's)
  • set the default: oto_identity(op='set', connector='grain', identity_id='<name>'); rename: op='rename' with new_name
  • ⚠️ no tool walks through the comptes on its own: a total across several comptes is obtained by calling each one (_account) and adding up

    • outils

    chargement…

    utilisation

    claude code — installe le plugin oto — claude plugin marketplace add otomata-tech/oto-plugin — mcp + skill configurés.
    claude.ai — branche mcp.oto.cx/mcp en connecteur custom (guide)
    cli — pipx install oto-cli puis oto grain …
    ← tous les connecteurscatalogue mcp →