Skip to content

tools.md

Latest commit

 

History

History
71 lines (44 loc) · 2.81 KB

tools.md

File metadata and controls

71 lines (44 loc) · 2.81 KB

Tools Guide

This guide covers the various tools and integrations available in the pipe project, including MCP and LSP servers.

Model Context Protocol (MCP) Server

The MCP server allows integration with AI agents for advanced tool calling capabilities.

Running the MCP Server

Docker

docker-compose up mcp

Manual

cd src
poetry run python -m pipe.cli.mcp_server

Integration with gemini-cli

To use advanced features like agent-driven Compression and session Forking in gemini-cli mode, you must first register pipe's tool server. This command tells gemini-cli how to communicate with this project's tools.

Execute the following command once:

gemini mcp add pipe_tools "python -m pipe.cli.mcp_server" --working-dir /path/to/pipe

(Replace /path/to/pipe with the actual absolute path to this project's directory.)

Benefit: After this integration, all tool calls made during a session (such as create_verified_summary or read_file) will become visible in the session history file (.json), providing complete transparency and auditability of the agent's actions.

Language Server Protocol (LSP) Server

pipe also provides a Language Server Protocol (LSP) server based on pygls. This server integrates with IDEs (like VS Code) to enable advanced understanding and manipulation of Python code. Its purpose is to provide context for LLMs to better understand the codebase.

Provided LSP Features:

  • Code Analysis: Provides definition locations and docstrings for classes, functions, and variables.
  • Code Snippets: Extracts code snippets for specific symbols.
  • Type Hints: Displays type hints for functions and classes.
  • Symbol References: Searches for symbol references within files.

How to Run

The LSP server can be started in the background using the following command. This allows IDEs to connect to the server.

python -m src.pipe.cli.pygls_server > /dev/null 2>&1 &
  • This command starts the server in the background and redirects standard output and standard error to /dev/null.
  • To stop the server, use the kill command with the Process Group ID (PGID) displayed upon startup. For example, if the PGID is 83272: 
    kill -s TERM -- -83272

Note: This server is designed to be connected to by an LSP client (e.g., an IDE like VS Code). Running it standalone will produce minimal visible output.

Tool Registration

When creating new tools, place the tool files in the appropriate directories. Tools are automatically registered if they follow the expected structure.

For Python-related tool development, you may find roles/python/craftsman.md helpful as it provides guidance on tool creation and best practices.

For more information on development, see docs/development.md.