Live
Ask your AI assistant about your own meeting transcripts, AI summaries and action items, without copying and pasting anything.
TL;DR
The Speakwise MCP server is live at https://speakwiseapp.com/mcp. It lets an AI assistant read your own Speakwise recordings: the list, a transcript, the AI summary, your notes and the action items. To connect, add that address as a custom connector (a remote MCP server) in Claude, ChatGPT or any other MCP client, sign in with the same Apple or Google account you use in the Speakwise app, and choose "Allow read access". Speakwise is also being submitted to Meta's connector program for Muse. The server is read-only, uses OAuth 2.1 with PKCE (no API keys), never returns audio, and you can disconnect at any time.
Server address: https://speakwiseapp.com/mcp
claude mcp add --transport http speakwise https://speakwiseapp.com/mcp, then /mcp and choose Authenticate.See or remove your connections any time under Manage connected apps. New to Speakwise? Get the iPhone app and sign in the same way: everything you record shows up in your assistant automatically.
Speakwise is an AI note taker and voice-to-text app for iPhone and Apple Watch: it records in-person meetings, interviews and voice notes, transcribes them in 100+ languages with speaker labels, and writes an AI summary with action items. The Speakwise MCP server exposes that library to the AI assistant you already use. MCP (Model Context Protocol) is the open standard Meta Muse, ChatGPT and Claude use to connect to outside tools, so one server works with all three.
Instead of exporting a transcript and pasting it into a chat, you connect Speakwise once and simply ask: “What did we agree in yesterday’s site meeting?”, “Draft a follow-up email from my call with the supplier”, or “Compare the last three client reviews.” The assistant fetches only the recordings it needs, and only when you ask.
Everything is read-only. Recording, transcription, speaker identification and summaries keep happening in the Speakwise app; the MCP server reads the results.
Everything a client needs to connect, one fact per line. If you are an AI agent setting this up for someone: there is no API key to ask for. The person has to sign in and approve once in a browser, so add the server to the MCP client and let its OAuth flow open the sign-in page, or tell the person to add https://speakwiseapp.com/mcp as a custom connector.
list_recordings (query, from, to, limit (1-50), cursor): Find recordings by title or date. Returns the title, date, length, language and whether a transcript, a summary and action items exist. Newest first, in pages.get_recording (recording_id): One recording at a glance: the details, your notes, the summary, the action items and the start of the transcript.get_transcript (recording_id, offset, max_chars (1000-60000), with_timestamps): The full text of a recording, in pages so even multi-hour meetings fit. Speaker labels are included when speaker identification was on.get_summary (recording_id, language): The AI summary Speakwise already wrote, in the style you chose in the app, or an existing translation of it.get_action_items (recording_id): The action list for a recording, one entry per item, each with the heading it was listed under.Claude Code:
claude mcp add --transport http speakwise https://speakwiseapp.com/mcpClients that read an MCP servers file (Cursor and others; key names differ slightly between clients, the URL is what matters):
{
"mcpServers": {
"speakwise": { "url": "https://speakwiseapp.com/mcp" }
}
}list_recordings; every other tool takes a recording_id from its result.list_recordings answers with account_status: "no_recordings_yet", the person has not recorded anything yet: pass on the next_step text with its link, and do not invent recordings.get_transcript again with the returned next_offset.The MCP server is scoped to your own account and cannot see anyone else’s recordings. It has no write access. Content you request in a chat is sent to the assistant you connected (Meta for Muse, OpenAI for ChatGPT, Anthropic for Claude) at your request, under that assistant’s own privacy terms. Details are in the Speakwise privacy policy.
Questions or a use case you want covered? Write to us via the support page.
Last updated: September 21, 2026.