MCP App Store
by KeenEthics
Granola icon

Granola MCP Server for Claude and ChatGPT

By Granola

Overview

Give Claude your Granola meeting context - notes, AI summaries and full transcripts. With Snapshots turned on, Claude can also see the slides, charts and designs shared in your meetings. Ask questions across your meetings and Claude finds the answer, citing the meetings it came from. Search by topic, person, company or date, browse your Granola folders, and pull exact quotes from transcripts. Try asking: • “What did we decide about pricing in last week’s meeting?” • “Where do things stand with Acme across all our calls?” • “Draft a follow-up email from today’s call with Acme.” • “What was on the roadmap slide in Tuesday’s planning meeting?” Claude only sees notes you can already access in Granola. Transcripts and full meeting history need a Business or Enterprise plan.

How to connect

  1. Claude: Open the Granola listing in the Claude connectors directory and choose Connect. Claude listing Granola exposes 22 tools here.
  2. ChatGPT: Open the Granola listing in the ChatGPT app directory and choose Connect. ChatGPT listing Granola exposes 21 tools here.

Tools

add_note_to_folder

ChatGPT
Add a Granola note to a folder. Use list_meeting_folders to find the destination folder ID and list_meetings to find the note ID; list_meetings shows which folders each note is already in. This operation is additive and idempotent. IMPORTANT: Adding a note grants everyone who can access the folder access to the note. A folder with visibility "workspace" shares the note with everyone in the workspace; check the folder's visibility and audience in list_meeting_folders first. Use only when the user has clearly requested that destination. The response confirms the write with an outcome (added or already_in_folder) and the destination folder's visibility and audience.

create_folder

ChatGPT
Create a top-level Granola folder. Use only when the user explicitly asks for a new folder; to file notes into an existing folder, find it with list_meeting_folders and use add_note_to_folder. visibility "private" (the default) creates a folder only the user can see. visibility "workspace" creates it in the workspace's Shared space, where everyone in the workspace can see every note added to it: use it only when the user asks for a team or shared folder. Sub-folders can't be created. If a top-level folder with the same title and visibility already exists, it is returned with outcome already_exists instead of creating a duplicate. The response has outcome created or already_exists and the folder with its id, visibility and audience; pass the id to add_note_to_folder.

create_meeting_note_from_transcript

ChatGPT
Create a private Granola meeting note from transcript text you already have, such as a meeting transcript from another app. Send the title, start time and whole transcript, in one call or, if it is long, in parts as described below. If neither the user nor the transcript says when the meeting took place, ask the user before calling; never guess a date or time. The note is created before this returns: url opens it right away, and meeting_id works with the other Granola tools. Granola then writes the note's summary in the background. The result's transcript_read_as lists how many utterances and which speakers were read; if that doesn't match the transcript, tell the user. Retrying with the same source, or without one the same speakers and words and meeting_started_at to the minute, returns the same note with outcome already_imported instead of a duplicate, even if the title or timestamps changed, and does not count again toward the daily import limit. Include source, attendees and meeting_platform whenever you have them. If you know the meeting's event in the user's Google or Outlook calendar, for example from a calendar tool or the meeting invite, pass it as calendar_event: a new note is then linked to that event and gets its attendees. Retrying an import that already succeeded returns the existing note unchanged, without linking it. Never guess it; if Granola can't find the event, the call fails with code calendar_event_unavailable and nothing is imported. Calls for the same user run one at a time, so concurrent calls wait rather than fail. If the user has reached the daily import limit, the error has code import_limit_reached: stop and tell the user instead of retrying. Send at most 100,000 characters, about 15,000 words, per call. Send a transcript longer than that, or one you aren't sure is shorter, in parts of up to that size split at line boundaries: the first with the meeting details and more_parts: true, later ones with only upload_id, transcript and more_parts: true, and the last without more_parts; the summary is written after the last part. A resent part adds nothing; an overlapping one is rejected with the time to continue from.

edit_voice_typing_dictionary

ChatGPT
Add or remove words and names in the user's voice typing (dictation) dictionary, so dictation spells them correctly. Add only names, product names and jargon that dictation gets wrong: not common words, and not every word from a document. Terms bias speech recognition, so ordinary words make dictation worse. Not used for meeting transcripts. Change only terms the user asked for or agreed to.

generate_summary

ChatGPT
Start generating a new summary for a Granola meeting the user owns, from its transcript and notes, only when the user explicitly asks. Existing summaries are kept; the new one is added alongside them. To use a template, find its ID with list_templates and pass it as template_id; otherwise Granola's standard summary is generated. For every meeting in a folder, find the folder with list_meeting_folders, discover its meetings with folder_id, and call this once per meeting. Generation runs in the background for about 30 to 90 seconds; this returns as soon as it has started. Read the meeting with get_meetings afterwards to see the new summary.

get_account_info

ChatGPT
Get the email, active workspace, joined workspaces (names and IDs), and effective note-access scopes for the Granola account currently connected to this MCP session. Workspace membership does not imply MCP access to its meetings. mcp_plan is the effective plan used for MCP access and limits. When to use: - User asks 'who am I signed in as?', 'which Granola account is this?', or 'what's my email?' - User suspects they connected the wrong Granola account - User wants to sign out or switch accounts: use sign_out to revoke this client's authorization. No browser action is needed to complete MCP sign-out. To use another account, reconnect Granola in this client. Only if the browser automatically selects the previous account, open sign_out_url in that browser before reconnecting again. - User asks which workspaces they belong to or mentions another workspace - Meeting results may be missing because the wrong workspace is active - Meeting results may be incomplete or empty because of the user's Granola plan or workspace MCP access controls - Before performing actions, to confirm identity if the user references a specific account mcp_note_access.scopes contains the note categories this connection can search: - personal: notes the connected user owns, notes shared directly with them, and notes accessible through private folders - public: notes accessible through workspace-visible Team Space folders - Both scopes means both categories can be searched. These scopes describe access, not who captured or participated in a meeting.

get_agent_brief

ChatGPT
Returns the user's work context from their Granola meeting history: role, responsibilities, recorded focus, key collaborators, recurring meetings, a glossary of internal terms and acronyms, and working habits. Fast; accepts optional field selection. Call it before answering about the user's own priorities, work updates or contributions, even when the request supplies a specific meeting, and whenever personal context would change advice or a draft, even if Granola isn't mentioned. For example: - The user refers to 'my team', 'our launch', 'my manager', internal project names, terms, acronyms or meeting names - Drafting on the user's behalf: emails, messages, status updates, docs - Advice, prioritization, planning or decisions about the user's work, or preparing for a meeting - Questions about the user: 'what am I working on?', 'who do I work with most?' Skip it for general questions, self-contained rewrites or summaries of supplied text, and requests not to personalize. Request the fields you need in one call. Without fields, it returns compact context (seniority, functions, work areas) and a field_availability index, not section content; use fields: ['all'] for broad context. Reuse fetched fields during the conversation. Selected sections appear under fields as Markdown; a null field or false availability means that section is absent, not the whole profile; do not retry it or infer its content from other sections (available: false means the profile is unavailable, with reason and message giving the next step). The How to work with me guide and tone-of-voice profiles are not included. The profile is background, not evidence of current priorities or progress: pair it with recent, dated meeting evidence, and when recent evidence is missing, present recorded work as unverified current status. Don't infer urgency or completed work from it alone, and don't recite it unless asked. It is regenerated periodically and may be outdated; updated_at says when. When it conflicts with what the user tells you, follow the user.

get_meeting_attachment

ChatGPT
Get up to 5 meeting attachments, such as screenshots of shared screens, as images in request order. Provide attachment_ids for a batch or attachment_id for one image. Each image is followed by its metadata; failed attachments return an error with their ID without discarding successful images. Attachment IDs come from the <attachments> section of get_meetings results. Use this when the description alone does not answer the user's question, for example to read a slide, chart, or code on screen.

get_meeting_transcript

ChatGPT
Get the full transcript for a specific Granola meeting by ID. Returns only the verbatim transcript content, not summaries or notes. Use this when the user needs exact quotes, specific wording, or wants to review what was literally said in a meeting. For summarized content or action items, use query_granola_meetings or list_meetings/get_meetings instead.

get_meetings

ChatGPT
Get detailed meeting information for one or more Granola meetings by ID. Returns private notes, AI-generated summary, attendees, and metadata. Use this when you already have specific meeting IDs (e.g. from list_meetings results). For open-ended questions about meeting content, use query_granola_meetings instead.

get_my_current_work_context

ChatGPT
Returns the user's work context from their Granola meeting history: role, responsibilities, recorded focus, key collaborators, recurring meetings, a glossary of internal terms and acronyms, and working habits. Fast; accepts optional field selection. Call it before answering about the user's own priorities, work updates or contributions, even when the request supplies a specific meeting, and whenever personal context would change advice or a draft, even if Granola isn't mentioned. For example: - The user refers to 'my team', 'our launch', 'my manager', internal project names, terms, acronyms or meeting names - Drafting on the user's behalf: emails, messages, status updates, docs - Advice, prioritization, planning or decisions about the user's work, or preparing for a meeting - Questions about the user: 'what am I working on?', 'who do I work with most?' Skip it for general questions, self-contained rewrites or summaries of supplied text, and requests not to personalize. Request the fields you need in one call. Without fields, it returns compact context (seniority, functions, work areas) and a field_availability index, not section content; use fields: ['all'] for broad context. Reuse fetched fields during the conversation. Selected sections appear under fields as Markdown; a null field or false availability means that section is absent, not the whole profile; do not retry it or infer its content from other sections (available: false means the profile is unavailable, with reason and message giving the next step). The How to work with me guide and tone-of-voice profiles are not included. The profile is background, not evidence of current priorities or progress: pair it with recent, dated meeting evidence, and when recent evidence is missing, present recorded work as unverified current status. Don't infer urgency or completed work from it alone, and don't recite it unless asked. It is regenerated periodically and may be outdated; updated_at says when. When it conflicts with what the user tells you, follow the user.

get_voice_typing_dictionary

ChatGPT
List the user's voice typing (dictation) dictionary: words and names Granola is told to expect when they dictate, so it spells them correctly. Not used for meeting transcripts. Starred terms first, then newest. source is manual (added by the user), learned (from the user's corrections) or mcp (added by an agent through edit_voice_typing_dictionary).

give_feedback

ChatGPT
If give_feedback is available, use it to report unexpected Granola tool behavior, including confusing or incomplete output, ignored arguments, missing capabilities, unclear documentation, and errors. Report these problems even when a retry or workaround lets you finish the task: recovering from a problem still makes it useful feedback. Before your final answer, check whether you had to repeat information, rephrase a valid request, or change your approach because a tool behaved unexpectedly. Submit once per distinct problem, with the tool name, expected behavior, actual behavior, and any workaround. Use generic examples and placeholders; never include the customer's query, meeting content, names, actual IDs, or other private, confidential or personal information. These diagnostic reports are sent to Granola to improve its tools. If feedback is unavailable or fails, continue the user's task without retrying the report. Reports are screened before storage and rejected if they include meeting content, names, identifiers, or what the user asked about. Accepted reports are stored by Granola under a 90-day retention policy. This tool does not provide support replies.

list_meeting_folders

ChatGPT
List the user's Granola meeting folders. Returns folder ID, title, description, note count including nested folders, visibility (private or workspace) and who can see the folder's notes. When to use: - User asks about their folders or wants to browse meetings by folder - User wants to narrow down meeting searches to a specific folder Use the returned folder IDs with list_meetings (folder_id parameter) to list meetings within a specific folder.

list_meetings

ChatGPT
List the user's Granola meeting notes within a time range. Returns meeting titles and metadata. IMPORTANT: For short-term questions about recent meeting details, prefer using query_granola_meetings instead. When to use: - User asks to list their meetings - User asks about action items, decisions, or summaries from meetings over a longer or specific date range - User asks about content from their meeting transcripts - User references 'Granola notes' or 'meeting notes' or 'transcripts' When NOT to use: - User is asking about upcoming calendar events or scheduling - User wants to create/modify calendar invites Use get_meetings to retrieve detailed meeting content after identifying relevant meetings.

list_templates

ChatGPT
List the note templates available for generating a meeting summary: Granola's built-in templates, the user's own, and those shared with their company, each with id, title, description, source (granola, yours or shared) and section headings. Pass a template's id as template_id to generate_summary; match the user's wording by title and description, and ask if more than one could fit.

query_granola_meetings

ChatGPT
Query Granola about the user's meetings using natural language. Returns a tailored response with inline citation links in mark (e.g. [[0]](url)) that reference source meeting notes. IMPORTANT: The response includes numbered citation links to specific Granola meeting notes. These citations MUST be preserved in your response to the user — they provide transparency and allow the user to verify information by clicking through to the original notes. When to use: - User asks about what was discussed, decided, or action-items from meetings - User asks about follow-ups, todos, or commitments from recent meetings - User references 'Granola notes' or 'meeting notes' When NOT to use: - User is asking about calendar scheduling or upcoming events - User explicitly asks for a specific meeting by ID (use get_meetings instead) Prioritize using query_granola_meetings over list_meetings/get_meetings for open-ended or natural language queries about meeting content.

remove_note_from_folder

ChatGPT
Remove a Granola note from a folder. Use list_meeting_folders to find the folder ID and list_meetings to find the note ID; list_meetings shows which folders each note is in. Removing the note may revoke folder-derived access for other people. The response confirms the write with an outcome (removed or not_in_folder) and the folder's visibility and audience.

search_meetings

ChatGPT
Find the user's meetings by keyword, participant/company, date, or a combination. Returns metadata only — title, date, participants and note headings, not what was said: read the relevant meetings with get_meetings before answering about their content. With keywords or participant, results are ordered by relevance, favouring recent meetings; searches cover the history allowed by your plan, narrowed by any supplied dates. With neither keywords nor participant, results are newest first; omitting both dates defaults to the last 7 days in the user's timezone. Results are paginated. Pass next_cursor as cursor to continue using the original filters. Omit filters or repeat the same values; limit and max_participants may change. Pages can be short or empty after access checks and recording deduplication; continue while next_cursor is present. completion_status is exhausted when no further matches remain, or incomplete if continuation is unavailable (including after 10,000 examined recordings); narrow the search in that case. Cursors expire after 15 minutes. Results reflect current access and indexed data, not a frozen snapshot. participant matches names, emails, companies, or domains in participant metadata, not mentions in notes. It is a hard AND filter alongside keywords, dates, and folder. Participant metadata can be incomplete and does not prove attendance. Use participant to find meetings with a person or company, keeping topic words in keywords. For broader questions about their work, account status, decisions, issue resolution, or meeting preparation, also search the person/company and topic in keywords with participant omitted: internal discussions may contain the answer. For example, pair participant='Acme', keywords=['onboarding'] with keywords=['Acme onboarding'] and no participant. Read relevant results from both searches before drawing a conclusion. Results include notes you can access. A note's creator may differ from the connected user. Captures of the same calendar event are grouped across pages, keeping the user's own note when they have one, otherwise the newest. duplicate_recordings can refer to a note returned on an earlier page. Use the listed note by default; read other_ids only for missing details needed to answer the question or an explicit request to compare attendees' notes. When to use: - User names an exact term, project, product or acronym and wants the meetings that mention it - User wants a list of matching meetings rather than an answer - User wants to browse meetings in a date range without searching for a word - User wants meetings with a named person or company, optionally about a specific topic - You need meeting ids to pass to another tool When NOT to use: - User wants the full notes of a meeting you already identified: use get_meetings Use attended_only=true for 'my meetings' or 'meetings I attended'; excludes upcoming calendar-only meetings. Use list_meeting_folders to discover folder IDs, then pass folder_id to search within a folder and its nested folders. Keyword matching is literal, so pass the words as the user said them plus obvious synonyms. Keywords must appear in the meeting title, summary, or the user's private notes; participant filters match participant metadata instead. Meetings with attachments you can view (shared-screen screenshots or photos the user attached) include an attachment_count attribute; a meeting without it has none you can view. Use get_meetings to see what each attachment is. Rate limits apply per user and per tool. Keep at most 2 search_meetings calls in flight at once; that is a ceiling, not a target, so fan out only for calls you already know you need. Any call can still get a rate_limit error: retry only failed calls, one at a time, after retry_after_seconds if given, else a short wait, and never when retryable is false.

search_transcripts

ChatGPT
Find specific words or quotes in accessible meeting transcripts in the active workspace. Returns ranked passages with approximate timestamps. Supply meeting_ids when known, batching meetings that share the same terms and filters. Matching is literal; put alternative words in one keywords array, e.g. ["pricing", "price", "cost"]. For what a person or company said, use participant (names/companies) or attendee_emails/attendee_domains (invite list). participant is an AND filter on metadata that can be incomplete, so for broader questions also search their name and the topic in keywords without it. Known meetings with no indexed match are also searched in their current transcripts. Without meeting_ids, search relies on an asynchronously updated index and groups duplicate recordings; empty results do not prove a topic was never discussed. Follow next_cursor even on empty pages, passing only cursor and optionally limit. If completion_status is incomplete, narrow the search or read known transcripts. Read surrounding dialogue with get_meeting_transcript using meeting_id, the passage's chunk_ids and transcript_version; adjacent chunk IDs continue, and omitting chunk_ids reads the whole conversation. counts is cumulative across pages; total_matching_chunks is a first-page estimate, null for direct search. Neither counts spoken mentions. Use another tool when: - The term is likely in the notes, or the user wants meetings about a topic: use search_meetings - You need the whole conversation: use get_meeting_transcript

sign_out

ChatGPT
Revoke the connected Granola account's consent and refresh tokens for this MCP client. Only use when the user asks to sign out, disconnect, or switch Granola accounts. This affects all connections for this account using the same client authorization. New MCP requests using this authorization will be rejected immediately after sign-out. No browser action is needed to complete MCP sign-out. If the user wants to switch accounts, tell them to reconnect Granola in this client. Only if the browser automatically selects their previous account, open the returned sign_out_url in that browser before reconnecting again.

switch_workspace

ChatGPT
Change the saved active Granola workspace, only when the user asks to switch: the default for the Granola app and every MCP client on the account, including other conversations and agents. Use get_account_info for workspace names and IDs. Later calls use the new default without a workspace argument. Meeting access and available tools depend on the workspace's plan and settings, so refresh the tool list after switching.

update_folder

ChatGPT
Rename a Granola folder or change its description. Use only when the user asks to rename or describe a folder. Get the folder ID from list_meeting_folders. Pass title, description or both; anything omitted stays the same. An empty description clears it. Requires owner or collaborator access to the folder (its role in list_meeting_folders). The response returns the updated folder, including its stored title and description.

update_meeting_title

ChatGPT
Rename a Granola meeting the user owns, only when the user asks. Find the meeting with search_meetings if you don't have its ID. Returns status updated, or unchanged if it already had this title. For an upcoming meeting on the user's calendar, pass its evt_ ID; if it has no note, this creates one first, only before it starts. The calendar event itself is not renamed.

update_private_notes

ChatGPT
Add content to the private notes of a Granola meeting the user owns (the notes they typed, not the AI summary), only when the user asks to add context or information. Existing notes are kept, and additions are stored as written by an agent so they stay distinguishable. Pass 1 to 20 operations, applied in order, each adding Markdown at the end (append) or start (prepend). Supported: paragraphs, and headings, bullet and numbered lists, unchecked checkboxes, bold, italic and links; tables, code, quotes, images and deeper headings are flattened or dropped. Read the notes with get_meetings first to avoid repeating what's there. To refresh the AI summary with the new notes, call generate_summary afterwards if it's available.

add_note_to_folder

Claude

create_folder

Claude

create_meeting_note_from_transcript

Claude

generate_summary

Claude

get_account_info

Claude

get_agent_brief

Claude

get_meeting_attachment

Claude

get_meeting_transcript

Claude

get_meetings

Claude

give_feedback

Claude

list_meeting_folders

Claude

list_meetings

Claude

list_templates

Claude

query_granola

Claude

query_granola_meetings

Claude

remove_note_from_folder

Claude

search_meetings

Claude

search_transcripts

Claude

sign_out

Claude

switch_workspace

Claude

update_folder

Claude

update_meeting_title

Claude

update_private_notes

Claude

update_summary

Claude

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/granola"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/granola)