tacit-mcp

MCP server that connects AI assistants to Tacit building digital twins. Ask questions about your buildings, equipment, sensors, and zones in natural language.

Works with Claude Desktop, Claude Code, Cursor, Windsurf, and any MCP-compatible client.

What it does

Four read-only tools:

The GraphQL tool includes the full schema reference, so the AI model can compose queries without needing separate documentation.

Quick start

Option A: npx (recommended — no install needed)

npx -y @tacit/mcp-server

Just point your MCP client at it (see configuration below). No cloning, no building.

Option B: Clone and build

git clone https://github.com/ucl-sbde/tacit-mcp.git
cd tacit-mcp
npm install
npm run build

You'll need a Tacit API key. Get one from your dashboard at app.betacit.com under Site Settings > API Keys.

Connection methods

1. Stdio transport (local, default)

The standard method — the MCP client launches the server as a child process. Best for individual use on your own machine.

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "tacit": {
      "command": "npx",
      "args": ["-y", "@tacit/mcp-server"],
      "env": {
        "TACIT_API_KEY": "your-api-key"
      }
    }
  }
}

Claude Code

Add to .mcp.json in your project:

{
  "mcpServers": {
    "tacit": {
      "command": "npx",
      "args": ["-y", "@tacit/mcp-server"],
      "env": {
        "TACIT_API_KEY": "your-api-key"
      }
    }
  }
}

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "tacit": {
      "command": "npx",
      "args": ["-y", "@tacit/mcp-server"],
      "env": {
        "TACIT_API_KEY": "your-api-key"
      }
    }
  }
}

2. Streamable HTTP transport (remote)

Run the server as a persistent HTTP service. Best for teams, cloud deployments, and environments where users can't install Node.js locally.

# Start the HTTP server
TACIT_API_KEY=your-api-key npm run start:http

# Or with npx
TACIT_API_KEY=your-api-key npx --package @tacit/mcp-server tacit-mcp-http

The server listens on http://0.0.0.0:3001/mcp by default.

Connect from any MCP client

Point your client at the server URL with a bearer token:

{
  "mcpServers": {
    "tacit": {
      "type": "streamable-http",
      "url": "https://your-host:3001/mcp",
      "headers": {
        "Authorization": "Bearer your-api-key"
      }
    }
  }
}

HTTP configuration

Health check

GET /health → { "status": "ok", "transport": "streamable-http", "sessions": 3 }

3. OAuth 2.1 (enterprise)

For production deployments where you want users to authenticate via Tacit's login flow instead of managing API keys:

TACIT_OAUTH_ISSUER=https://app.betacit.com npm run start:http

This enables:

• Dynamic client registration — MCP clients register automatically

• Authorization code + PKCE — users log in through Tacit's web UI

• Token refresh — sessions stay alive without re-authentication

• Token revocation — clean session termination

MCP clients that support OAuth (like Claude Desktop) will discover the auth configuration automatically via the .well-known/oauth-authorization-server metadata endpoint.

4. Docker

docker run -p 3001:3001 -e TACIT_API_KEY=your-api-key tacit/mcp-server

Connect using the HTTP transport config above.

Try it

Once connected, ask your AI assistant things like:

• "List all my building sites"

• "What AHUs are in Tower West?"

• "Show me temperature sensors on AHU-001"

• "Get the last 24 hours of supply air temperature data"

• "What equipment feeds the lobby zone?"

Environment variables

Development

npm run dev       # watch mode — stdio transport
npm run dev:http  # watch mode — HTTP transport
npm run build     # compile TypeScript
npm start         # run stdio transport
npm run start:http # run HTTP transport

License

MIT