deepset

deepset-mcp is the MCP server for the Deepset AI platform. It exposes over 30 specialized tools to help AI agents build, debug, and orchestrate pipelines against the Haystack Enterprise AI platform. In practice, you instantiate the MCP server in your environment (via uv) and then load the server into Cursor or other MCP clients to access the available tools and platform resources through a Python SDK. The included wrapper and environment variables let you connect to your workspace and authenticate requests programmatically. Use the provided 30+ tools to query, transform, run, and debug pipeline components, and leverage the Python SDK for scripting against the Haystack Enterprise platform.

Prerequisites:

• Python 3.8+ and pip
• uv Python package (for running MCP servers via uv)
• Access credentials for the Deepset Haystack Enterprise platform (workspace and API key)

Installation steps:

1. Install uv and set up the project dependencies using the recommended workflow:

# Install uv as described in the project docs
pipx install uv

# Install project dependencies and extras (if applicable in your environment)
uv sync --locked --all-extras --all-groups

2. Prepare the local development script if you want to run the MCP server directly via a wrapper (as shown in the repository):

# Example: use a helper script to run the local deepset-mcp server
# Ensure the run-deepset-mcp.sh is executable:
chmod +x run-deepset-mcp.sh

3. Run the MCP server through your MCP client configuration (as described in the mcp_config section):

# This uses uvx to run the local package (adjust package name as appropriate in your environment)
uvx deepset-mcp

4. Validate the server is reachable and the tools are listed via your MCP client or the documentation endpoints.

Note: If you modify the codebase, restart the MCP server to pick up changes.

Environment variables frequently required by the Deepset MCP server include DEEPSET_WORKSPACE and DEEPSET_API_KEY for authenticating with the Haystack Enterprise platform. If you encounter connectivity or authentication issues, double-check that the workspace value matches your Deepset account and that the API key has the proper scopes. When developing locally, use the provided wrapper script to avoid PATH dependencies, and remember to restart the MCP server after code changes. The Makefile targets mentioned in the documentation (lint, format, types, test) help maintain code quality and ensure compatibility with type checks and unit/integration tests. If you run into environment-specific issues, confirm that uv and Python dependencies are installed and that the virtual environment contains the deepset-mcp entry point.