About Recount AI Notetaker
Recount MCP server connects the local desktop meeting recorder Recount to AI apps, for people who need to search meeting content and organize follow-up tasks. Recording and transcription happen on the computer, and the server relays requests over a local address.
Key features
- 01Search transcripts across all meetings
- 02View the live transcript of a meeting
- 03Start, stop, or mute recordings
- 04Read and save meeting notes
- 05Manage transcript speaker labels
- 06Create and manage tasks and projects
Use cases
- 01Find decisions or details from past meetings
- 02Follow an ongoing meeting
- 03Turn meeting action items into board tasks
- 04Import recordings and create meeting records
MCP support
The MCP server exposes tools to AI clients for listing meetings, searching transcripts, controlling recordings, editing notes, labeling speakers, and managing a task board. It communicates with Recount on the local computer through 127.0.0.1 and does not relay messages elsewhere.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
Tools
Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.
Details
From the project READMERecount MCP server
Connect Claude, Cursor, Codex, Gemini CLI, GitHub Copilot and other AI apps to Recount AI Notetaker, the desktop app that records meetings on your own computer and transcribes them there, with no bot joining the call.
Download Recount free for Windows, macOS or Linux, then connect your AI app below.
Once connected, your AI app can:
- answer questions about past meetings ("What did Sarah say about the budget last week?")
- search what was said across every meeting
- follow the live transcript of the meeting you're in
- start, stop and mute recording
- name the speakers in a transcript
- write your notes and meeting notes
- turn action items into tasks on your Recount task board
Everything stays on your computer. Recount serves your meetings to AI apps only on 127.0.0.1 (this computer), and this package passes messages between your AI app and Recount without sending them anywhere else. Your AI app itself will of course read what it asks for, under its own privacy terms.
Requirements
- Recount for Windows, macOS or Linux (free), open and signed in
- Node.js 18 or later, for the
npxsetups below
Connecting AI apps is free in Recount. Recount transcribes English only.
Setup
Pick your AI app. Recount's Connect AI page shows the same steps.
Claude Desktop
Easiest: download recount.mcpb from the latest release and double-click it. Or click Connect on Recount's Connect AI page.
Or add this to claude_desktop_config.json (Settings > Developer > Edit Config):
{
"mcpServers": {
"recount": {
"command": "npx",
"args": ["-y", "@externalsoftware/recount-mcp"]
}
}
}
Claude Code
claude mcp add --transport http --scope user recount http://127.0.0.1:7891/mcp
Cursor
In ~/.cursor/mcp.json:
{
"mcpServers": {
"recount": { "url": "http://127.0.0.1:7891/mcp" }
}
}
VS Code (GitHub Copilot)
In your user mcp.json (Command Palette > MCP: Open User Configuration):
{
"servers": {
"recount": { "type": "http", "url": "http://127.0.0.1:7891/mcp" }
}
}
Windsurf
In ~/.codeium/windsurf/mcp_config.json:
{
"mcpServers": {
"recount": { "serverUrl": "http://127.0.0.1:7891/mcp" }
}
}
Gemini CLI
In ~/.gemini/settings.json:
{
"mcpServers": {
"recount": { "httpUrl": "http://127.0.0.1:7891/mcp" }
}
}
OpenAI Codex
In ~/.codex/config.toml:
[mcp_servers.recount]
url = "http://127.0.0.1:7891/mcp"
Any other MCP app
Apps that connect to a URL: http://127.0.0.1:7891/mcp (Streamable HTTP). Apps that start a program: command npx, arguments -y @externalsoftware/recount-mcp.
Tools
| Tool | What it does |
|---|---|
list_meetings | Recorded meetings, newest first, with who was in each |
get_meeting | One meeting: details, voices, transcript, your notes, meeting notes and screenshots |
search_transcripts | Find what was said across every meeting |
get_live_transcript | What has been said so far in the meeting being recorded |
get_recording_status | Whether Recount is recording right now, and for how long |
start_recording / stop_recording | Start or stop recording a meeting (microphone and computer audio) |
mute_recording | Mute or unmute the microphone or computer audio while recording |
import_recording | Import an audio or video file (.m4a, .mp3, .wav, .mp4...) as a new meeting |
get_my_notes / write_my_notes | Read or write your own notes for a meeting |
save_meeting_notes | Save meeting notes (Markdown) on a meeting |
list_note_styles / get_note_style | The meeting-note styles you can choose from |
rename_meeting / reprocess_meeting / delete_meeting | Rename, transcribe again, or delete a meeting |
name_speaker / confirm_speaker / ignore_speaker / reset_speaker | Label the voices in a meeting's transcript |
list_participants / add_participant / update_participant / remove_participant | The people listed as being in a meeting |
list_tasks / get_task / create_task / update_task / delete_task | Your task board |
add_task_column / create_project / update_project | Board columns and projects |
Naming a speaker only labels that voice in that meeting's transcript; Recount doesn't save anyone's voice from it.
Troubleshooting
- "Recount isn't open": open Recount and sign in. AI apps can only reach it while it's running.
- Locked: if Recount's window is locked, it refuses AI requests until you unlock it.
- Different port: if you changed Recount's port, run
npx -y @externalsoftware/recount-mcp --port=1234or setRECOUNT_MCP_URL.
Development
npm test # tests against a stand-in Recount server
node build-mcpb.mjs # builds dist/recount.mcpb for Claude Desktop
License
MIT for this package. Recount itself is free to download, with an optional Pro subscription: https://externalsoftware.com/recount/?utm_source=recount_mcp&utm_medium=referral
Made by External Software LLC. Recount was called Hark before November 2026.
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.