MCP server and A2A agent for the Recordist app (https://recordist.app): lets your AI assistant read and act on the meetings recorded on your own computer. Listens on loopback by default.
- ✓Open-source license (MIT)
- ✓Actively maintained (<30d)
- ✓Clear description
- ✓Topics declared
- ✓Documented (README)
claude mcp add gateway -- npx -y @recordist/gateway{
"mcpServers": {
"gateway": {
"command": "npx",
"args": ["-y", "@recordist/gateway"]
}
}
}MCP Servers overview
# @recordist/gateway
[](https://mcpservers.org/servers/recordist-app/gateway) [](https://glama.ai/mcp/servers/recordist-app/gateway)
MCP server + A2A agent for **Recordist** (https://recordist.app), the app that
writes notes from your calls on your own computer, with no bot in the call. It lets
Claude Desktop, Claude Code, Cursor and other agents read and act on the meetings
recorded on your machine.
Recordist itself is opening in small waves: ask for a place at
**https://recordist.app/early-access** (Mac with Apple silicon for now).
- **Reads** (list, get, transcript, search, action items) work whether or not the
Recordist app is running: the gateway talks to the app's local API when it is
up and falls back to opening `recordist.db` read-only when it is not.
- **Actions** (start and stop recording, add a marker, regenerate notes) need the
app. When it is not running they fail with `Recordist app is not running`.
`start_recording` also needs **Allow agents to start recordings** in Recordist's
Settings → Integrations, which is off by default.
- **Where your meetings go.** The gateway itself talks only to the Recordist app on
`127.0.0.1` and listens on loopback by default. What it returns goes to the AI
assistant you connect, which may send it to its own cloud model.
`regenerate_notes` has the app write notes with the provider set in Recordist's
Settings → AI notes, which may also be a cloud service.
## Install
```bash
npx -y @recordist/gateway --doctor # check data dir, DB, token, app
```
Requires Node 20+. `better-sqlite3` ships prebuilt binaries for common platforms.
## Connect an assistant
### Claude Desktop
`claude_desktop_config.json` (macOS: `~/Library/Application Support/Claude/`,
Windows: `%APPDATA%\Claude\`):
```json
{
"mcpServers": {
"recordist": {
"command": "npx",
"args": ["-y", "@recordist/gateway"]
}
}
}
```
### Claude Code
```bash
claude mcp add recordist -- npx -y @recordist/gateway
```
or, for the HTTP transport while `npx -y @recordist/gateway --http` is running (it needs
the app's API token, see [Authentication](#authentication)):
```bash
TOKEN="$(cat ~/Library/Application\ Support/app.recordist.desktop/api_token)"
claude mcp add --transport http recordist http://127.0.0.1:47322/mcp \
--header "Authorization: Bearer $TOKEN"
```
### Cursor
`.cursor/mcp.json` (project) or `~/.cursor/mcp.json` (global):
```json
{
"mcpServers": {
"recordist": {
"command": "npx",
"args": ["-y", "@recordist/gateway"]
}
}
}
```
### Any Streamable HTTP client
```bash
npx -y @recordist/gateway --http # http://127.0.0.1:47322/mcp
```
Send `Authorization: Bearer <api_token>` with every request.
### Authentication
The HTTP transport (`--http`) and the A2A agent (`--a2a`) answer only requests
that carry `Authorization: Bearer <api_token>`, where the token is the contents
of `<data>/api_token` (or `RECORDIST_API_TOKEN`), the same token the gateway
uses for the app's local API. A missing or wrong token gets `401` with
`WWW-Authenticate: Bearer` and nothing else. The one exception is the A2A Agent
Card, which stays public so clients can discover the scheme; it lists skills
only, never meetings. Neither mode starts when no token is available, and
neither does `--all`, stdio included: open Recordist once so it writes
`api_token`, or run with no flags for stdio alone. The stdio transport needs no
token, since only the assistant that launched it can talk to it. `--doctor` says
whether `--http` and `--a2a` will require the token or refuse to start.
## What the assistant gets
**Tools**
| Tool | Purpose |
|---|---|
| `list_meetings` | Recent meetings; optional `q`, `from`/`to`, `limit`/`offset` |
| `get_meeting` | One meeting with notes, action items and markers |
| `get_transcript` | Transcript as `json` \| `md` \| `srt` \| `vtt` \| `txt` |
| `search_meetings` | Full-text search across transcripts and notes (FTS5) |
| `get_action_items` | Action items, filter by `meeting_id` / `open_only` |
| `regenerate_notes` | Re-run AI notes (`summary`, `action_items`, `decisions`, …) with the provider set in Recordist; needs the app |
| `start_recording` | Start recording; needs the app and **Allow agents to start recordings** (Settings → Integrations, off by default) |
| `stop_recording` | Stop the current recording; needs the app |
| `add_marker` | Bookmark the current moment; needs the app |
**Resources**: `recordist://meeting/{id}` (JSON) and
`recordist://meeting/{id}/transcript` (Markdown). The resource list shows the
50 most recent meetings.
**Prompts**: `summarize_meeting`, `draft_followup_email`, `weekly_review`.
## A2A agent
```bash
npx -y @recordist/gateway --a2a # http://127.0.0.1:47323/
```
Agent Card at `http://127.0.0.1:47323/.well-known/agent.json` (public, no
token; `authentication.schemes` is `["bearer"]`); skills mirror the tools above.
JSON-RPC 2.0 at `/` with `tasks/send`, `tasks/get`, `tasks/cancel` and
`tasks/sendSubscribe` (SSE); every JSON-RPC request needs
`Authorization: Bearer <api_token>`.
Natural-language tasks are routed to a skill with a small keyword router; every
completed task returns a **text** part and a **data** part with the structured
result. To skip the router, send a data part `{"skill": "...", "args": {...}}`.
```bash
TOKEN="$(cat ~/Library/Application\ Support/app.recordist.desktop/api_token)"
curl -s http://127.0.0.1:47323/ \
-H "Authorization: Bearer $TOKEN" \
-H 'content-type: application/json' \
-d '{
"jsonrpc": "2.0", "id": 1, "method": "tasks/send",
"params": {
"id": "task-1",
"message": { "role": "user", "parts": [
{ "type": "text", "text": "What are my open action items from this week?" }
]}
}
}' | jq .result.artifacts[0].parts
# explicit skill call
curl -s http://127.0.0.1:47323/ -H "Authorization: Bearer $TOKEN" -H 'content-type: application/json' -d '{
"jsonrpc":"2.0","id":2,"method":"tasks/send",
"params":{"message":{"role":"user","parts":[
{"type":"data","data":{"skill":"get_transcript","args":{"meeting_id":"01J…","format":"srt"}}}
]}}}'
# streaming
curl -N http://127.0.0.1:47323/ -H "Authorization: Bearer $TOKEN" -H 'content-type: application/json' -d '{
"jsonrpc":"2.0","id":3,"method":"tasks/sendSubscribe",
"params":{"message":{"role":"user","parts":[{"type":"text","text":"list my meetings from today"}]}}}'
```
## CLI
```
npx -y @recordist/gateway MCP over stdio (default)
npx -y @recordist/gateway --http MCP over Streamable HTTP on 127.0.0.1:47322/mcp (bearer token)
npx -y @recordist/gateway --a2a A2A agent on 127.0.0.1:47323 (bearer token)
npx -y @recordist/gateway --all stdio + --http + --a2a (exits if there is no token)
npx -y @recordist/gateway --doctor diagnostics: data dir, DB, token, app reachability, HTTP/A2A auth
npx -y @recordist/gateway --version
--http-port <n> --a2a-port <n> --host <addr>
```
Always run it with npx by its scoped name, `@recordist/gateway`. A global
install (`npm i -g @recordist/gateway`) puts the same command on your path as
`recordist-gateway`.
`--host` binds `--http` and `--a2a` to another address and exists for
containers. Do not expose it on a network interface: the token would cross the
network in plain HTTP.
Logs go to stderr; stdout is reserved for the stdio MCP transport.
## Environment variables
| Variable | Default | Purpose |
|---|---|---|
| `RECORDIST_DATA_DIR` | macOS `~/Library/Application Support/app.recordist.desktop`<br>Windows `%APPDATA%\app.recordist.desktop`<br>Linux `~/.local/share/app.recordist.desktop` | Where `recordist.db` and `api_token` live |
| `RECORDIST_API_URL` | `http://127.0.0.1:47321` | Local API base URL |
| `RECORDIST_API_TOKEN` | contents of `<data>/api_token` | Bearer token for the local API, and the token `--http` and `--a2a` require from callers |
## How backend selection works
On each call the gateway probes `GET /v1/health` (800 ms timeout, result cached
for 5 s). If the app answers, the call goes to the API. Otherwise reads open
`<data>/recordist.db` with `readonly: true` (the app's WAL is never written to)
and actions return `Recordist app is not running`. FTS5 (`segments_fts`) is used
for search when present, with a `LIKE` fallback otherwise.
## Security notes
- The gateway binds `127.0.0.1` by default. `--host` exists for containers; do
not expose it on a network interface. The Streamable HTTP transport enables
DNS-rebinding protection and only accepts its own Host (`127.0.0.1` or
`localhost` by default). The A2A agent accepts only its own Host, no browser
Origin unless allow-listed, and JSON-RPC POSTs only as `application/json`.
- The gateway itself talks only to the Recordist app on `127.0.0.1`. What it
returns goes to the AI assistant you connect, which may send it to its own
cloud model, so connect only an assistant you trust with your meetings.
- `--http` and `--a2a` require `Authorization: Bearer <api_token>` on every
request (the A2A Agent Card excepted), compared in constant time, so other
user accounts on the machine cannot read meetings through them. Programs
running under your own account can read `api_token`, so the token does not
separate them from you.
- The API token is read from `<data>/api_token`, which the app writes with
mode `0600`. `--doctor` warns if the file is group/world readable. Never commit
or share it; anyone with the token can control recording on your machine.
- The SQLite database is opened read-only; the gateway never modifies it.
- No telemetry. The gateway makes no network calls other than to the Recordist
app's local API.
## Changes in 0.1.7
- Security: dependency updates for advisories in @modelcontextprotocol/sdk and
transitive packages. The SDK floor iWhat people ask about gateway
What is recordist-app/gateway?
+
recordist-app/gateway is mcp servers for the Claude AI ecosystem. MCP server and A2A agent for the Recordist app (https://recordist.app): lets your AI assistant read and act on the meetings recorded on your own computer. Listens on loopback by default. It has 0 GitHub stars and its last recorded update is dated 2026-10-12.
How do I install gateway?
+
You can install gateway by cloning the repository (https://github.com/recordist-app/gateway) or following the README instructions on GitHub. ClaudeWave also provides quick install blocks on this page.
Is recordist-app/gateway safe to use?
+
Our security agent has analyzed recordist-app/gateway and assigned a Trust Score of 95/100 (tier: Verified). See the full breakdown of passed checks and flags on this page.
Who maintains recordist-app/gateway?
+
recordist-app/gateway is maintained by recordist-app. The last recorded GitHub activity is dated 2026-10-12, with 3 open issues.
Are there alternatives to gateway?
+
Yes. On ClaudeWave you can browse similar mcp servers at /categories/mcp, sorted by popularity or recent activity.
Deploy gateway to your cloud
Ship this repo to production in minutes. Each platform spins up its own environment with editable env vars.
Maintain this repo? Add a badge to your README
Drop the badge into your GitHub README to show it's tracked on ClaudeWave. Each badge links back to this page and reflects the live Trust Score.
[](https://claudewave.com/repo/recordist-app-gateway)<a href="https://claudewave.com/repo/recordist-app-gateway"><img src="https://claudewave.com/api/badge/recordist-app-gateway" alt="Featured on ClaudeWave: recordist-app/gateway" width="320" height="64" /></a>More MCP Servers
Fair-code workflow automation platform with native AI capabilities. Combine visual building with custom code, self-host or cloud, 400+ integrations.
Reverse engineer anything with agents, from app behavior down to native binaries.
User-friendly AI Interface (Supports Ollama, OpenAI API, ...)
An open-source AI agent that brings the power of Gemini directly into your terminal.
Real-time global intelligence dashboard. AI-powered news aggregation, geopolitical monitoring, and infrastructure tracking in a unified situational awareness interface
🕷️ An adaptive Web Scraping framework that handles everything from a single request to a full-scale crawl! Don't be shy, join here: https://discord.gg/EMgGbDceNQ and follow here for daily tips and tricks: https://x.com/Scrapling_dev