# The MCP server

The archive handed to agents as Model Context Protocol tools.



Fluncle serves a small, stateless [Model Context Protocol](https://modelcontextprotocol.io) server: the same archive the public API exposes, handed to agents as tools. Streamable HTTP, no sessions, no auth.

```
https://www.fluncle.com/mcp
```

## Connect [#connect]

Point any MCP client at the endpoint. For a client that reads a config block:

```json
{
  "mcpServers": {
    "fluncle": { "type": "http", "url": "https://www.fluncle.com/mcp" }
  }
}
```

## Tools [#tools]

The archive, as tools:

| Tool                   | Does                                                    |
| ---------------------- | ------------------------------------------------------- |
| `list_findings`        | The latest findings and mixtapes, newest found first.   |
| `list_tracks`          | Every track Fluncle holds, newest release first, paged. |
| `get_random_track`     | One finding, pulled at random.                          |
| `search_tracks`        | Search Spotify for a candidate.                         |
| `submit_track`         | Recommend a track for review (not a publish).           |
| `subscribe_newsletter` | Board the mothership.                                   |

`submit_track` is a recommendation, like the [CLI](/docs/cli)'s submit: it queues a track, and Fluncle listens before anything joins the archive.

## How it's built [#how-its-built]

The MCP server is a thin layer over the same internal functions the [`/api`](/docs/api-overview) routes use, so validation, the submission rate limit, and the submitter hash all stay identical to every other surface. An agent and a human reach the same archive, the same way.

For discovery, the MCP Server Card lives at [`/.well-known/mcp/server-card.json`](https://www.fluncle.com/.well-known/mcp/server-card.json).