file-tools-mcp: Instructions file for Claude Code

CLAUDE.md

please follow the coding guidance mentioned in ~/.claude/docs/writing-code.md Follow additional guidance mentioned in ~/.claude/docs/python.md & ~/.claude/docs/using-uv.md

Use the available MCP tools for searching up the latest documentation on MCP server implementation and follow the best practices outlined there. We will be using the official python-sdk for MCP servers. Use context7 mcp server to get the latest documentation and examples when implementing the server.

Project Overview

This is an MCP (Model Context Protocol) server implementation for file tools. The goal is provide tools to LLMs and coding agents to quickly get context of a project by getting project structure and contents of multiple files in a single request.

Available Tools

project_structure

Generates a tree view of a directory structure while respecting .gitignore patterns.

Parameters:

• path (str, optional): Directory path to analyze (default: current working directory)
• max_depth (int, optional): Maximum depth to traverse (default: None for unlimited)
• show_hidden (bool, optional): Include hidden files and directories (default: False)

Example usage:

# Get structure of current directory
result = await session.call_tool("project_structure", {})

# Get structure with depth limit
result = await session.call_tool("project_structure", {
    "path": "/path/to/project",
    "max_depth": 3
})

# Include hidden files
result = await session.call_tool("project_structure", {
    "show_hidden": True
})

read_multiple_files

Reads multiple files and returns their contents in a single response with clear separation.

Parameters:

• paths (list[str], required): List of file paths to read (absolute or relative paths)

Features:

• Each file's content is clearly separated with headers showing the file path and size
• Handles errors gracefully (file not found, permission denied, binary files)
• Returns all results in a single response for efficient LLM processing