Otter.ai MCP Server
Official Otter.ai MCP server. Lets AI assistants search meeting transcripts, retrieve summaries, surface action items, and analyze conversation history.
Otter.ai's MCP server is an OAuth-authenticated remote connector that gives MCP-compatible AI clients (Claude, ChatGPT, and others) secure access to a user's Otter.ai meeting library. Instead of copy-pasting transcripts or exporting files, an AI assistant can search across all of a user's recorded meetings, retrieve full transcripts, pull AI-generated summaries, and extract action items directly inside the chat session.
The server is part of Otter's broader Conversational Knowledge Engine, which positions Otter as both an MCP client (pulling context from Gmail, Google Drive, Notion, Jira, and Salesforce) and an MCP server (pushing meeting context out to other AI tools). Access is OAuth based with granular permissions, so the AI assistant can only read meetings the user has explicitly authorized.
Notably, this is a provider-hosted remote MCP server, not a self-hosted package. There is no GitHub repo to clone or npm package to install. Clients connect to the hosted endpoint at https://mcp.otter.ai/mcp and complete an OAuth flow. Several community-built alternatives exist on GitHub (for example, DarrenZal/otter-mcp), but those are unofficial and use email/password authentication against the private API.
Tools
| Tool | Description |
|---|---|
search_transcripts |
Search across the user's authorized Otter conversations by keyword, topic, participant, or phrase. Searches transcript content, not just titles and summaries. |
list_meetings |
List recent meetings with metadata such as title, date, duration, and participants. |
get_meeting_summary |
Retrieve the AI-generated summary and action items for a specific meeting. |
get_transcript |
Return the full text transcript of a specific Otter conversation. |
Prerequisites
- An Otter.ai account with access to the MCP Server feature (admin may need to enable it for your workspace).
- An MCP-compatible client. The Otter MCP server is listed in the Claude connector directory and works with ChatGPT and other remote-MCP clients.
Connect from Claude (recommended)
- Open Claude on the web (
claude.ai). Direct setup from the Claude desktop app is not currently supported, but once connected on the web the connector appears in the desktop app automatically. - Go to Settings → Connectors → Browse connectors.
- Find Otter.ai in the directory and click Connect.
- Complete the OAuth flow in the Otter window and authorize the meetings you want Claude to access.
Connect a custom MCP client
For clients that accept a remote MCP URL directly, point them at:
https://mcp.otter.ai/mcp
Example custom-connector config (Claude Desktop claude_desktop_config.json style for remote MCP):
{
"mcpServers": {
"otter": {
"url": "https://mcp.otter.ai/mcp"
}
}
}
The first request will trigger an OAuth handshake in the browser. After authorization, the client stores a token and can call the Otter tools.
Notes
- All access is OAuth-authenticated with granular, user-scoped permissions.
- The MCP feature may require an Otter Business or Enterprise plan; confirm with your workspace admin or Otter support if the connector is not visible.
- Ask Claude to summarize themes across the last quarter of customer discovery interviews recorded in Otter, without manually exporting transcripts.
- Have an AI assistant pull action items from yesterday's standup and draft follow-up emails or Jira tickets.
- Search a library of sales calls for every mention of a competitor or feature request, then cluster the results.
- Generate a recap document or board update by combining Otter meeting summaries with notes from other connected tools.
- Cross-reference commitments made on calls against a product spec or PRD by querying transcripts in natural language.
- "Search my Otter meetings from the last 30 days for any discussion of pricing pushback and summarize the objections."
- "List my meetings with Acme Corp this quarter and give me the action items from each one."
- "Pull the full transcript of yesterday's leadership sync and draft a recap email for the team."
- "Across all customer interviews tagged this month, what are the top three feature requests?"
- "Find the meeting where we discussed the Q3 roadmap and extract every commitment made by engineering."
- Official, provider-hosted server: no self-hosting, no community fork, no scraping the private API.
- OAuth with granular per-meeting permissions, so users control exactly what the AI can see.
- Works out of the box in Claude via the connector directory, plus any remote-MCP client via the public URL.
- Bidirectional integration story: Otter's own AI Chat acts as an MCP client to Gmail, Drive, Notion, Jira, and Salesforce.
- The MCP feature appears to be gated to Otter Business / Enterprise plans and may require workspace admin provisioning.
- No public, self-hostable open-source server; you depend on Otter's hosted endpoint and its uptime.
- The exact published tool schema is not fully documented publicly; available tools are described in help-center and blog material rather than a developer reference.
- DarrenZal/otter-mcp (community, open-source): self-hosted Python MCP server that authenticates with Otter email/password and exposes search and transcript retrieval tools.
- bcharleson/otter-cli (community): an agent-native CLI plus MCP server built on the Otter.ai public API (requires Otter Enterprise API access).
- Zapier MCP for Otter.ai: routes Otter actions through Zapier's MCP endpoint, useful if you also want Zapier's wider app catalog.