> ## Documentation Index
> Fetch the complete documentation index at: https://docs.orinadus.com/llms.txt
> Use this file to discover all available pages before exploring further.

# urchin mcp

> Start the MCP server over stdio for IDE and agent integrations.

## Synopsis

```bash theme={null}
urchin mcp
```

No flags. The server starts on stdin/stdout and runs until the parent process closes the pipe.

## Protocol

JSON-RPC 2.0 over stdio. Each message is a newline-terminated JSON object. The server speaks the [Model Context Protocol](https://modelcontextprotocol.io) — IDEs that support MCP (Cursor, Zed, VS Code with the Copilot extension) can connect by launching this command as a subprocess.

## Tools exposed

The server exposes 10 tools. See [MCP Overview](/urchin/api/mcp-overview) for the full list.

## IDE configuration

<Tabs>
  <Tab title="VS Code / Copilot Chat">
    Create `.vscode/mcp.json` in your workspace root:

    ```json theme={null}
    {
      "servers": {
        "urchin": {
          "type": "stdio",
          "command": "urchin",
          "args": ["mcp"]
        }
      }
    }
    ```

    Restart VS Code. The MCP tools appear in Copilot Chat automatically.
  </Tab>

  <Tab title="Cursor">
    Create or update `.cursor/mcp.json` at the repo root:

    ```json theme={null}
    {
      "mcpServers": {
        "urchin": {
          "command": "urchin",
          "args": ["mcp"]
        }
      }
    }
    ```

    Cursor picks this up when you open the folder.
  </Tab>

  <Tab title="Zed">
    Add to `~/.config/zed/settings.json`:

    ```json theme={null}
    {
      "context_servers": {
        "urchin": {
          "command": {
            "path": "urchin",
            "args": ["mcp"]
          }
        }
      }
    }
    ```
  </Tab>
</Tabs>

## Requirements

`urchin` must be on `PATH`. Install via:

```bash theme={null}
cargo install --path crates/urchin-cli
```

Or add `~/.cargo/bin` to your `PATH` if you built locally.

## Notes

* The MCP server does not require `urchin serve` to be running. It reads and writes the journal file directly.
* Each tool call reads the journal from disk. There is no in-memory cache between calls.
* If the journal does not exist, query tools return empty results (not an error).
* Errors from tool calls return `isError: true` in the MCP response envelope.