mcp-claude-config-analyser-v2

Claude Anthropic MCP Servers Tool Analyzer

This project provides tools for extracting, formatting, and analyzing MCP (Model Context Protocol) server tools. It consists of three main components:

1. analyser.py: Extracts tool information from MCP servers
2. extract_tools.py: Formats the extracted tool data in various formats
3. reporter.py: Analyzes tool descriptions for potential issues

Installation

Prerequisites

• Python 3.8 or higher
• pip package manager

Setup

1. Clone this repository:

   git clone https://github.com/kubiosec-ai/mcp-claude-config-analyser-v2.git
   cd mcp-claude-config-analyser-v2
   

2. Create a python environment

   python3 -m venv .venv
   

   source .venv/bin/activate
   

3. Install required dependencies:

   pip install -r requirements.txt
   

4. Set OpenAI API key

   export OPENAI_API_KEY=xxxxxxxxxxx
   

5. Make a backup of your Claude Desktop Config file

   cp /Users/xxradar/Library/Application\ Support/Claude/claude_desktop_config.json ./claude_desktop_config.json
   

Usage

1. Extract Tool Information from MCP Servers

Use analyser.py to connect to MCP servers and extract their tool lists:

python analyser.py [options]

Options:

• --dry-run: Parse config without connecting to servers
• --server SERVER: Process only a specific server
• --timeout TIMEOUT: Connection timeout in seconds (default: 30)
• --config CONFIG: Path to Anthropic Claude config file (default: ./claude_desktop_config.json)
• --output OUTPUT: Path to output file (default: config.json)

Example:

# Extract tools from all servers with a 20-second timeout
python analyser.py --timeout 20

# Extract tools from a specific server
python analyser.py --server MCP_DOCKER

# Dry run to test configuration parsing
python analyser.py --dry-run

2. Format Tool Data

Use extract_tools.py to format the extracted tool data in various formats:

python extract_tools.py [options]

Options:

• --config CONFIG: Path to the config file (default: config.json)
• --output OUTPUT: Path to the output file (default: tool_list.json)
• --format FORMAT: Output format (default: json)
  • json: Standard JSON format with server_name, tool_name, and description fields
  • csv: CSV format with server_name, tool_name, and description columns
  • reporter: Special format compatible with reporter.py

Example:

# Generate a standard JSON file
python extract_tools.py --output tool_list.json

# Generate a CSV file
python extract_tools.py --format csv --output tool_list.csv

# Generate a reporter-compatible JSON file
python extract_tools.py --format reporter --output reporter_tools.json

3. Analyze Tool Descriptions

Use reporter.py to analyze tool descriptions for potential issues:

python reporter.py <json-file>

Example:

python reporter.py tool_list.json

Output Formats

1. Standard JSON Format

[
  {
    "server_name": "playwright",
    "tool_name": "browser_close",
    "description": "Close the page"
  },
  ...
]

2. CSV Format

server_name,tool_name,description
playwright,browser_close,Close the page
...

3. Reporter Format

{
  "tools": [
    {
      "name": "browser_close",
      "description": "Close the page"
    },
    ...
  ]
}

Common Workflows

Default

# Extract tools from all servers
python analyser.py --timeout 30

# Format tools 
python extract_tools.py --output tool_list.json

# Analyze tool descriptions
python reporter.py tool_list.json

Extract and Analyze All Tools

# Extract tools from all servers
python analyser.py --timeout 30

# Format tools for reporter
python extract_tools.py --format reporter --output reporter_tools.json

# Analyze tool descriptions
python reporter.py reporter_tools.json

Extract Tools from a Specific Server

# Extract tools from a specific server
python analyser.py --server MCP_DOCKER

# Format tools as CSV
python extract_tools.py --format csv --output mcp_docker_tools.csv

Experimental OpenAI SDK Agent

python extract_tools.py --format reporter --output reporter_tools.json

python new_reporter.py 

Troubleshooting

Connection Issues

If you encounter connection issues with MCP servers:

1. Increase the timeout value: --timeout 60
2. Try connecting to servers individually: --server SERVER_NAME
3. Check if the server is running and accessible

Parsing Errors

If you encounter JSON parsing errors:

1. Use the --dry-run option to test configuration parsing
2. Check the format of your config.json file
3. Ensure the MCP server is returning valid JSON responses

License

MIT License