MCPStack

MCPStack is a Python-based orchestrator that lets you build and run pipelines of MCP tools in a scikit-learn-like workflow for LLMs. You can stack multiple MCP tools into a single pipeline, expose only the tools you want the LLM to access, and execute the pipeline in convenient environments such as Claude Desktop or through programmatic Python APIs. The CLI (mcpstack) provides utilities to list available tools, inspect presets (preconfigured pipelines), create new pipelines, configure individual tools, and run pipelines with a chosen configuration. Tools are auto-registered via entry points, so once MCPStack is installed, you can add desired tools to your environment and MCPStack will recognize them automatically. The workflow supports both interactive usage and programmatic composition, enabling reproducible experiments with presets or custom tool chains.

Prerequisites:

• Python 3.9 or newer
• Either uv (recommended) or pip for installation

Installation options:

1. Using UV (recommended):

uv add mcpstack

2. Using pip:

pip install mcpstack

Post-installation steps (optional but recommended):

• Install pre-commit hooks to ensure code quality:
  • UV: uv run pre-commit install
  • Pip: pre-commit install

Notes:

• MCPStack is the orchestrator that discovers MCP tools via project.entry-points.["mcpstack.tools"] and auto-registers them when installed in your environment.
• You can install additional MCP tools into the same environment to expand the available tooling in your pipelines.

Tips and tips-to-remember:

• The CLI workflow revolves around mcpstack executed via uv run mcpstack (or globally installed mcpstack).
• Use tools <tool_name> configure to customize a tool before adding it to a pipeline, and then pipeline to assemble or run it.
• Presets offer ready-to-run pipelines to reproduce experiments quickly.
• If you encounter environment issues, ensure your Python environment has access to your MCP tools' dependencies and that entry points are correctly registered.
• For reproducible results, save pipelines with the programmatic API or via CLI, then re-run with the same configuration.
• MCPStack emphasizes security by letting you expose only the tools you include in your pipeline; this constrains the LLM's tool access accordingly.