Granola

Metadata	Value
Category	productivity
Capabilities	http
Website	https://granola.ai

Returns shapes

• conversation — from op_get_conversation
• conversation[] — from op_list_conversations
• meeting — from op_get_meeting
• meeting[] — from op_list_meetings

Connections

• api — Granola REST API — live data, requires network
• cache — Local app cache — instant, offline, read-only

Readme

Meeting transcripts and AI summaries from Granola — automatically captured as you meet.

Setup

Granola must be installed and have run at least once. No API key needed — auth is read directly from Granola’s local token file.

If you see auth errors, open the Granola app to refresh the token.

Connections (API + cache)

The skill declares two connections — api (live REST) and cache (local cache-v6.json). For operations that support both, pick one with params.connection: "api" or "cache". The default is api (first in the list).

Connection	When	Use case
api	Live calls with token	Freshest data, full transcripts
cache	Reads local cache file	Fast, works offline

get_meeting uses the api connection only (the cache does not store full transcript text).

What gets created in the graph

For each get_meeting call:

Entity	Type	Details
The meeting	meeting	Title, times, location, AI summary as description
Transcript	transcript	Full text body (FTS5-indexed), segment count, duration

Relationships:

• transcript --transcribe--> meeting

Attendees are stored in data.attendees on the meeting entity. Full person entity creation from attendees is a future enhancement.

Usage

list_meetings — Browse recent meetings

curl -X POST http://localhost:3456/api/skills/granola/meeting.list \
  -H "X-Agent: cursor" \
  -H "Content-Type: application/json" \
  -d '{"limit": 10}'

Returns meetings with title, times, location, attendees. No transcript data — use get_meeting for that.

get_meeting — Full meeting with transcript

curl -X POST http://localhost:3456/api/skills/granola/meeting.get \
  -H "X-Agent: cursor" \
  -H "Content-Type: application/json" \
  -d '{"id": "2fbc5ac2-..."}'

Returns the full meeting including:

• Transcript as a linked transcript entity (FTS5-indexed)
• AI summary as the meeting description
• All attendees with enrichment (name, avatar, job title)

Transcript format

Transcripts are stored as plain text in the transcript entity body:

[00:02:34] You: What's the status on the deployment?
[00:02:41] Other: We hit a snag with the Docker config, but it's fixed now.
[00:02:58] You: Great. What's the timeline?

Speaker labels: You (microphone) and Other (system audio).

Q&A conversations

Granola lets you chat with AI about meeting transcripts. Each meeting can have one or more Q&A threads.

What this is on the graph

Each thread is a conversation entity — the same entity kind used elsewhere on the graph for chat threads and ordered message histories (email threads, assistant chats, SMS-style threads, etc.). AgentOS cares what it is (a named thread with messages and a URL), not which product produced it. list_conversations returns thin rows (title, ids, notes_url); get_conversation fills text / content by joining message bodies. The document_id you pass is the meeting document id in Granola; treat it as “which meeting this Q&A is about” when reasoning, even though the remembered entity type for the thread itself is still conversation.

Workflow: Find AI chats about a meeting

1. Get the meeting id — list_meetings (or use the id from a meeting summary)
2. List threads — list_conversations(document_id: meeting_id) → returns thread(s) with titles and ids
3. Read a thread — get_conversation(thread_id: thread_id) → full user/assistant message history

Via AgentOS MCP:

run({ skill: "granola", tool: "list_meetings", params: { limit: 10 } })
run({ skill: "granola", tool: "list_conversations", params: { document_id: "<id from step 1>" } })
run({ skill: "granola", tool: "get_conversation", params: { thread_id: "<id from step 2>" } })

list_conversations — Q&A threads for a meeting

curl -X POST http://localhost:3456/api/skills/granola/conversation.list \
  -H "X-Agent: cursor" \
  -H "Content-Type: application/json" \
  -d '{"document_id": "6dc09094-1c7e-421d-86c1-2b23f924b34e"}'

Returns threads linked to that meeting (e.g. “Validator Agent and Auto Rewrite Loop”). Use get_conversation with a thread ID to read the full exchange.

get_conversation — Full Q&A thread

curl -X POST http://localhost:3456/api/skills/granola/conversation.get \
  -H "X-Agent: cursor" \
  -H "Content-Type: application/json" \
  -d '{"thread_id": "5c4516ae-1224-4e39-a642-a1b9b7e0e279"}'

Returns the thread with all user/assistant messages in order.