Skip to content
🚀 AI App Builders

Build full-stack apps from natural language

LovableReplitBolt.newBase44V0 by VercelBrowse All →
💻 AI Coding Tools

Code faster with AI pair programmers

GitHub CopilotCursorWindsurfTabnineCodexClaude CodeBrowse All →
🔍 AI Research Tools

Find and synthesize research faster

PerplexityChatGPTConsensusElicitBrowse All →
🎙️ AI Meeting Assistants

Record, transcribe, and summarize meetings

Fireflies.aiFathomOtter.aiGranolaBrowse All →

AI Note Takers With MCP: Permission Scope Compared

By Etienne Tawong

Vendor facts verified

Compare what Granola, Fireflies, Otter, Fathom, Avoma, Fellow, tl;dv, and Read AI document about MCP access to meeting notes: permission scope, authorization, setup, and named AI assistants.

MCP access compared from official documentation

What each vendor documents about MCP access to meeting data
ToolWhat the connection can reachAuthorization and setup documentedAI assistants namedOfficial source
GranolaSet by the Granola plan: Basic and Free reach your own notes from the last 30 days; Business adds your full personal history, workspace notes, transcripts and write tools; on Enterprise, access is admin-configurable and MCP is off by default.See official sourceSee official sourceGranola MCP
FirefliesYour Fireflies meetings, queried from a connected assistant.OAuthClaude, ChatGPTFireflies MCP server guide
OtterMeetings you explicitly authorize.OAuthClaude, ChatGPTOtter MCP Server article
FathomMeetings you already have permission to view in Fathom.See official sourceClaude, ChatGPTFathom MCP article
AvomaMeetings the signed-in user has permission to view. Also documents three write operations: meeting purpose, outcome, and privacy.OAuth 2.0; an admin can publish the connector for the whole organizationClaude, ChatGPTAvoma MCP page
FellowData you already have permission to view in Fellow.OAuth; a workspace admin must first allow MCP connectionsClaude and ChatGPT (custom connectors), CursorFellow MCP server reference
tl;dvMeeting lists, metadata, transcripts, and highlights.tl;dv API key; server run with Docker or Node.js; Business or Enterprise tl;dv account (per April 2025 announcement)Claude Desktop, Cursor, other MCP clientstl;dv MCP announcement
Read AIYour Read AI meetings; search covers roughly the last 12 months. Includes a tool that can share a meeting report with another person.OAuth 2.1 with your Read AI account; open betaChatGPT and Codex, Claude, Microsoft Copilot Studio, GammaRead AI MCP Server article

This table lists documented scope only. "See official source" means the cited page does not state that detail plainly enough for us to summarize it. We list plan requirements only where the vendor states them, as Granola and tl;dv do. DecideNavigator has not run its own MCP test of these tools.

Granola: access depends on your plan

Granola documents that its MCP connection reaches different notes by plan. Basic and Free cover your own notes from the last 30 days. Business adds your full personal history, workspace notes, transcripts and write tools. On Enterprise, MCP is off by default and admins decide who gets access and what agents can do. Check the plan your workspace holds before assuming older or shared notes will be available.

Granola MCP plan access

Fireflies, Otter, and Fathom

Fireflies documents OAuth access to your Fireflies meetings from Claude or ChatGPT. Otter documents OAuth and limits the connection to meetings you explicitly authorize. Fathom limits MCP access to meetings you already have permission to view. In each case the account you sign in with determines what the assistant can reach.

Otter MCP Server article

Avoma and Fellow: admin controls

Avoma scopes MCP data to the signed-in user's access level and says an admin can publish the connector once for the whole organization. Unlike a read-only connector, Avoma documents write operations that change meeting purpose, outcome, and privacy, so decide who may use them. Fellow requires a workspace admin to allow MCP connections first and says connections are logged in workspace security settings.

Fellow MCP server reference

tl;dv and Read AI: different setup models

tl;dv announced a developer-run server in April 2025 that uses a tl;dv API key and runs with Docker or Node.js, and said it needs a Business or Enterprise account. Read AI offers a hosted server in open beta with OAuth 2.1 sign-in. Read AI also documents a tool that can share a meeting report with another person, so review that action before enabling it.

Read AI MCP Server article

Host requirements are separate from the note taker plan

Two products must allow the connection. The note taker controls which meetings are exposed and may tie that to its own plan or admin setting. The AI assistant you connect it to, such as Claude or ChatGPT, has its own plan, connector, and admin settings. Check both sides with the vendor documentation and your administrator. This guide does not state which assistant plans support custom connectors.

How to evaluate permission scope

Before connecting, list what the assistant should be able to reach: only your meetings, shared workspace meetings, or meetings selected one by one. Compare that with the table. Then confirm who can enable or disable the connector, how access is revoked, whether admins can restrict it, and whether the connector can change records or share them, not only read them.

How to check answers and citations with a sample meeting

If you decide to try a connection, use a sample meeting your organization has approved for testing and that contains no sensitive information. Ask a few questions with known answers, such as a decision, an owner, and a date. Check whether each answer points back to the right meeting and whether the assistant can reach any meeting outside the scope you expected. This is a suggested procedure for readers, not a test DecideNavigator has run.

Related reading