概要
What is mcpx-py?
mcpx-py is a Python library for interacting with large language models (LLMs) using tools provided by mcp.run. It supports all models supported by PydanticAI and offers both a Python API and a command-line interface.
How to use mcpx-py?
Install the library via uv add mcpx-py or pip install mcpx-py. Obtain an mcp.run session ID by running npx --yes -p @dylibso/mcpx gen-session --write and set the MCP_RUN_SESSION_ID environment variable. Then import mcpx_py to use the Chat class or invoke the mcpx-client command-line tool.
Key features of mcpx-py
- Python library for LLM interaction via mcp.run tools.
- Supports all models supported by PydanticAI.
- Structured output with Pydantic models.
- Command line interface for chat, list, and tool calls.
- Works with Anthropic, OpenAI, Gemini, Ollama, and Llamafile.
Use cases of mcpx-py
- Summarize web content using an LLM.
- Call remote tools via mcp.run from Python or CLI.
- Build AI applications with structured data outputs.
- Experiment with multiple LLM providers.
FAQ from mcpx-py
What dependencies are required?
You need uv, npm, and optionally ollama for local models.
How do I set up authentication?
Generate an mcp.run session ID using npx --yes -p @dylibso/mcpx gen-session and either write it to a config file (with --write) or set the MCP_RUN_SESSION_ID environment variable.
Which LLM providers are supported?
The library supports Claude, OpenAI, Gemini, Ollama, Llamafile, and any model supported by PydanticAI.
How do I get structured output from the LLM?
Pass a Pydantic BaseModel subclass as the result_type parameter when creating a Chat instance, then call send_message_sync.
Can I use mcpx-py without installing it permanently?
Yes, you can run mcpx-client directly using uvx --from mcpx-py mcpx-client without permanent installation.
