fabric

The Fabric MCP Server acts as a bridge between MCP-enabled hosts (such as IDE extensions or chat interfaces) and a running Fabric REST API powered by fabric --serve. It translates MCP requests into REST calls to your Fabric instance and streams responses back to the MCP client. Using this server, MCP hosts can discover available Fabric patterns and models, request pattern runs, fetch configuration details, and retrieve Fabric-related data through standard MCP tooling semantics. The design preserves Fabric’s workflow ergonomics while enabling integration into broader MCP-enabled toolchains.

From an MCP client, you can discover tools exposed by the server (e.g., list_tools to enumerate Fabric patterns, list patterns/details, run a chosen pattern, list models/strategies, retrieve configuration). When you invoke a tool such as running a Fabric pattern, the MCP Host forwards the request to the Fabric MCP Server, which then calls the corresponding Fabric REST API (via fabric --serve). The server streams the Fabric response back as an MCP response, allowing interactive or streaming results to be consumed by your IDE, chat, or other MCP clients.

Prerequisites:

• Python >= 3.10
• Optional: uv for development (uv is a Python environment manager) or use a virtual environment

Installation from PyPI (end users):

pip install fabric-mcp

If you prefer using uv and a development workflow:

# Install uv if not installed
pip install uv

# Install the package in a per-project virtual environment (via uv helper commands)
# The following demonstrates installing the package in the active environment
uv pip install fabric-mcp

Run the server (after installation):

# Default stdio transport
fabric-mcp --stdio

# Or enable HTTP streamable transport if your client requires HTTP
fabric-mcp --http-streamable --host 0.0.0.0 --port 8000 --mcp-path /message

If you prefer running via pipx (isolated CLI):

pipx install fabric-mcp
pipx run fabric-mcp

Environment setup (optional):

• Ensure Fabric REST API is reachable and configured to accept requests from this MCP server
• Set environment variables as needed (see Configuration section below)

Configuration tips:

• FABRIC_BASE_URL should point to your running Fabric REST API (default http://127.0.0.1:8080)
• FABRIC_API_KEY is required only if your Fabric REST API enforces authentication
• FABRIC_MCP_LOG_LEVEL controls the verbosity of the fabric-mcp server (DEBUG, INFO, WARNING, ERROR, CRITICAL)

Transport options:

• --stdio is the default transport (standard I/O)
• --http-streamable enables an HTTP server for MCP communications; you can customize host, port, and path via --host, --port, and --mcp-path

Common issues:

• Ensure the Fabric REST API is reachable from the MCP server host and that API keys (if used) are correct
• If using HTTP transport, ensure firewall rules allow traffic to the chosen port
• When upgrading fabric-mcp, reinstall in your environment to refresh dependencies