litemcp

litemcp is a lightweight MCP server client designed to help you rapidly integrate AI SDKs (such as LangChain, OpenAI Agent SDK, and direct OpenAI API usage) into your MCP projects. It exposes tool-access APIs that let you bind available tools to your LLM runtimes, run agent workflows, and execute tool calls via a simple, minimal interface. You can use MCPServerManager to start the server locally, then obtain specialized tool collections (e.g., agent SDK tools, LangChain-compatible tools) and feed them into your agents or chat wrappers. This enables you to integrate external SDKs and custom tool implementations without heavy boilerplate, while keeping dependencies small and your integration logic straightforward. The library emphasizes safe tool usage, including optional validators to guard against undesired tool calls and human-in-the-loop approvals when needed.

Prerequisites:

• Python 3.8+ (recommended) -pip (comes with Python)
• Access to install Python packages from PyPI

Install the MCP server package:

pip install litemcp

Verify installation (example):

python -m litemcp --help

Run the MCP server (example, assuming the package provides a CLI entry point compatible with -m usage):

python -m litemcp

If you prefer to install and run via an isolated environment, you can use a virtual environment:

python -m venv venv
source venv/bin/activate  # on Windows use: venv\Scripts\activate
pip install litemcp
python -m litemcp

Notes and tips:

• The MCP configuration supports enabling/disabling servers and excluding tools; adjust mcpServers accordingly in your config file.
• If you integrate with tools like LangChain, you can bind the MCP-provided tools to your chain object and execute tool calls as part of the chain flow.
• For safety, consider enabling a validator function to approve or block specific tool calls before they execute.
• Ensure your Python environment has network access to fetch any required SDKs or APIs when binding to external tools.
• Check for compatibility between your chosen AI runtime (e.g., OpenAI SDK, LangChain) and the MCP tools you expose; keep tool interfaces consistent to avoid runtime errors.
• If you encounter issues with tool invocation, inspect tool_call payloads and the corresponding tool implementations to verify name resolution and argument handling.