# Phonotheca MCP server

Remote MCP server for [Phonotheca](https://phonotheca.com), the archive for
recorded interviews. Connect Claude or any MCP client to your workspace and
search every transcript, browse projects and interviews, and read
speaker-labelled transcripts with timestamps. Passages come back with their
speaker, interview, project, and timestamp attached, so a quote arrives with
its citation.

The connection is read-only. It cannot upload audio, start transcription,
edit a transcript, or delete anything.

## Endpoint

```
https://phonotheca.com/api/mcp
```

Streamable HTTP, stateless. Authentication is a bearer token created in
[Settings](https://phonotheca.com/app/settings), shown once and revocable
there at any time. A free account is enough to try it; the archive, search,
and export stay open permanently.

## Connect

Claude Code:

```sh
claude mcp add --transport http phonotheca \
  https://phonotheca.com/api/mcp \
  --header "Authorization: Bearer YOUR_TOKEN"
```

Claude API:

```json
"mcp_servers": [{
  "type": "url",
  "url": "https://phonotheca.com/api/mcp",
  "name": "phonotheca",
  "authorization_token": "YOUR_TOKEN"
}]
```

## Tools

| Tool | What it does |
|---|---|
| `search_transcripts` | Full-text search across every transcript, optional project filter. Returns cited passages. |
| `list_projects` | The workspace's projects with interview counts. |
| `list_interviews` | A project's interviews with duration, language, and status. |
| `get_transcript` | A speaker-labelled transcript with timestamps, paged for long interviews. |

## Data handling

Requests hit the same EU-resident application and database as the workspace.
Transcript text returned to an MCP client goes to whatever assistant you
connected, under that assistant's terms. Revoking the token ends access
immediately. Details: [phonotheca.com/mcp](https://phonotheca.com/mcp).