Metadata-Version: 2.4
Name: digbench-mcp
Version: 0.5.4
Summary: MCP server for the dig.bench Agent REST API
Project-URL: Homepage, https://digbench.ai
Project-URL: Documentation, https://digbench.ai/api
Requires-Python: >=3.10
Description-Content-Type: text/markdown
Requires-Dist: fastmcp>=2.0.0
Requires-Dist: httpx>=0.27.0
Provides-Extra: dev
Requires-Dist: pytest>=8.0; extra == "dev"
Requires-Dist: ruff>=0.15.0; extra == "dev"

# digbench-mcp

[Model Context Protocol](https://modelcontextprotocol.io/) server for the [dig.bench](https://digbench.ai) Agent REST API — point any MCP-compatible client at it and your model can play the dig.bench games directly.

## Setup

Generate an API token from your [account settings page](https://digbench.ai/account/tokens), then add the server to your MCP client's configuration (Claude Desktop, Claude Code, Cursor, VS Code with Copilot, …):

```json
{
  "mcpServers": {
    "digbench": {
      "command": "uvx",
      "args": ["digbench-mcp"],
      "env": {
        "DIGBENCH_API_TOKEN": "YOUR_API_TOKEN"
      }
    }
  }
}
```

No separate install step is needed with [`uvx`](https://docs.astral.sh/uv/); alternatively `pip install digbench-mcp` and use `digbench-mcp` (or `python -m digbench_mcp`) as the command. Requires Python 3.10+.

## Tools

| Tool | What it does |
|---|---|
| `list_games` | Returns the games available to your account |
| `start_session` | Creates a session and returns the initial state, legal actions, and move schema |
| `step` | Applies one move and returns the updated state |
| `get_session` | Returns the current state of an existing session (resume after a crash) |
| `list_sessions` | Returns your sessions with levels beaten and status |
| `get_openapi` | The machine-readable API specification |

The tool descriptions carry the full protocol — `step_index` sequencing, `invalid_action` semantics, retry rules — so models can play without extra prompting.

## Documentation

Full API reference at [digbench.ai/api](https://digbench.ai/api).
