GC-Eval MCP Server

A manifest-driven MCP server that exposes BizChat Gov Cloud quality evaluation data to any MCP-compatible client (e.g., Copilot CLI).

No repo access required. Install once, always up-to-date — data is fetched from a hosted endpoint at runtime.

Quick Start (for consumers)

pip install gc-eval-mcp

Add to .copilot/mcp.json in any project:

{
  "mcpServers": {
    "GC-Eval": {
      "command": "gc-eval-mcp"
    }
  }
}

Done. Restart Copilot CLI and the tools are available.

How It Works

┌─────────────────────────────────────┐
│  Hosted Data (GitHub Pages / Blob)  │
│  manifest.json → describes datasets │
│  ├── HTML reports                   │
│  ├── JSON files                     │
│  ├── Markdown docs                  │
│  └── SQLite databases               │
└──────────────┬──────────────────────┘
               │ fetched at runtime
┌──────────────▼──────────────────────┐
│  GC-Eval MCP Server (local)         │
│  Reads manifest → registers tools   │
│  Caches data locally (~/.cache/)    │
└──────────────┬──────────────────────┘
               │ MCP protocol
┌──────────────▼──────────────────────┐
│  Copilot CLI / Any MCP Client       │
└─────────────────────────────────────┘

Available Tools

Always available (static)

Registered per dataset (dynamic, from manifest)

Default tools (from example manifest)

Configuration

Adding New Data (for maintainers)

To expose new data through the MCP, update the manifest — no server code changes needed:

{
  "id": "my-new-data",
  "type": "sqlite",
  "tool_prefix": "my_data",
  "path": "data/my_new_data.db",
  "description": "My new dataset"
}

Supported types: html-collection, markdown, json, sqlite.

Upload the data file to the hosted endpoint, add the entry to manifest.json, and all MCP users instantly have access.

Development

cd mcp-server
pip install -e .
gc-eval-mcp

Or run directly:

python -m gc_eval_mcp

Manifest Schema

See manifest.json for the full example. Key fields:

{
  "schema_version": "1.0",
  "data_version": "2026-05-07",
  "base_url": "https://your-host.com/data/",
  "datasets": [
    {
      "id": "unique-id",
      "type": "html-collection|markdown|json|sqlite",
      "tool_prefix": "valid_python_identifier",
      "path": "relative/path/to/data",
      "index": "optional/index.json",
      "description": "Human-readable description"
    }
  ]
}