hayhooks

Hayhooks is a Python-based MCP server that deploys Haystack pipelines and agents as REST APIs and exposes them as MCP Tools for use in AI development environments. After installation, you start the server with hayhooks run, which launches a configurable HTTP API surface for each deployed pipeline or agent. The server supports Open WebUI streaming backends, an embedded Chainlit UI, and open MCP endpoints so you can interact with your Haystack components from tools like Cursor, Claude Desktop, or other MCP clients. Use it to deploy, undeploy, list, or run pipelines and agents through natural MCP tool invocations or direct HTTP requests. The quick start shows how to install Hayhooks, start the server, and deploy a minimal agent wrapper that exposes a /my_agent/run endpoint as well as a streaming /chat/completions interface, illustrating how Hayhooks bridges Haystack with MCP-based tooling.

Prerequisites:

• Python 3.8+ installed on your system
• pip available in your PATH
• Internet access to install dependencies

Step 1: Create a Python environment (optional but recommended)

• macOS/Linux: python3 -m venv hayhooks-env source hayhooks-env/bin/activate
• Windows: python -m venv hayhooks-env hayhooks-env\Scripts\activate

Step 2: Install Hayhooks

pip install hayhooks

Step 3: Run Hayhooks (MCP server)

hayhooks run

Step 4: Deploy a pipeline or agent (example shown in docs)

• Prepare a wrapper dir as described in the Hayhooks quick start, then deploy via the Hayhooks CLI, for example:

hayhooks pipeline deploy-files -n my_agent ./my_agent_dir

Step 5: Access the server

• REST API endpoint (default port 1416): http://localhost:1416/my_agent/run
• MCP tooling endpoints are exposed for integration with MCP clients.

Note: If you prefer using a Docker-based workflow, you can run Hayhooks in a container and map port 1416 to your host. See the project documentation for Docker usage and additional configuration options.

Tips and caveats:

• The default port is 1416; adjust the PORT environment variable in mcp_config if you need a different port.
• Hayhooks supports both YAML-based and wrapper-based Haystack pipeline deployments; refer to the documentation for details on both approaches.
• When integrating with MCP clients (e.g., Cursor, Claude Desktop), expose each pipeline or agent as an MCP Tool via Hayhooks so they show up as usable tools in your MCP environment.
• For streaming OpenAI-compatible chat interfaces, you can leverage the /chat/completions endpoint with streaming enabled as demonstrated in the quick start example.
• If you encounter issues, enable verbose logging in Hayhooks and check for Python dependency compatibility with your Haystack version.
• To uninstall or upgrade, use pip uninstall hayhooks or pip install --upgrade hayhooks respectively.