Code Summarizer

A command-line tool that summarizes code files in a given directory using Gemini Flash 2.0. Now with MCP server support for integration with LLM tools!

Features

• Recursively processes code files in a directory
• Respects .gitignore rules
• Skips irrelevant directories like node_modules, dist, etc.
• Summarizes code files using Gemini Flash 2.0
• Outputs summaries to a text file
• Configurable detail level and summary length
• MCP server for integration with Claude Desktop and other LLM tools
• Modular design for easy integration into other applications
• Secure API key management
• Authentication for MCP server endpoints
• Retry mechanism with exponential backoff for LLM calls
• Rate limiting to prevent abuse

MCP Server Integration Details

MCP Resources

• code://file/* - Access individual code files
• code://directory/* - List code files in a directory
• summary://file/* - Get summary for a specific file
• summary://batch/* - Get summaries for multiple files

MCP Tools

• summarize_file - Summarize a single file with options
• summarize_directory - Summarize a directory with options
• set_config - Update configuration options

MCP Prompts

• code_summary - Prompt template for summarizing code
• directory_summary - Prompt template for summarizing entire directories

Options

• --detail, -d: Set the level of detail for summaries. Options are 'low', 'medium', or 'high'. Default is 'medium'.
• --max-length, -l: Maximum length of each summary in characters. Default is 500.

Security Features

API Key Management

• API keys are stored securely and prioritize environment variables over configuration files
• Keys are validated for proper format before use
• API keys are never exposed in logs or error messages
• Configuration file doesn't store API keys when they're provided via environment variables

Authentication

• All MCP server endpoints (except health check) require authentication via API key
• Authentication uses the x-api-key header for secure transmission
• Failed authentication attempts are logged for security monitoring

Rate Limiting

• Built-in rate limiting prevents abuse of the service
• Default: 60 requests per minute per IP address
• Configurable through server settings

Error Handling

• Structured error system with categorization
• Sensitive information is never exposed in error messages
• Proper error codes are returned for different failure scenarios

LLM Call Resilience

• Automatic retry with exponential backoff for transient failures
• Configurable retry settings including max retries, delays, and backoff factor
• Jitter added to retry timing to prevent thundering herd problems
• Request ID tracking for tracing issues across the system

Supported File Types

• TypeScript (.ts, .tsx)
• JavaScript (.js, .jsx)
• Python (.py)
• Java (.java)
• C++ (.cpp)
• C (.c)
• Go (.go)
• Ruby (.rb)
• PHP (.php)
• C# (.cs)
• Swift (.swift)
• Rust (.rs)
• Kotlin (.kt)
• Scala (.scala)
• Vue (.vue)
• HTML (.html)
• CSS (.css, .scss, .less)

How It Works

1. The tool scans the specified directory recursively, respecting .gitignore rules.
2. It filters files based on supported extensions.
3. For each supported file, it reads the content and determines the programming language.
4. It sends the code to Gemini Flash 2.0 with a prompt to summarize, including detail level and length constraints.
5. The summaries are collected and written to the specified output file.

Output Format

The output file will have the following format:

relative/path/to/file
Summary text here

relative/path/to/next/file
Next summary text here

Project Structure

• index.ts: Main CLI implementation
• src/: Source code directory
  • summarizer/: Core summarization functionality
  • mcp/: MCP server implementation
  • config/: Configuration management
• bin/: CLI entrypoint
• config.json: Default configuration file
• tsconfig.json: TypeScript configuration
• package.json: Project dependencies and scripts
• .env.example: Template for setting up environment variables
• .gitignore: Files and directories to ignore in Git
• __tests__: Unit and integration tests
• __mocks__/mock-codebase: Mock codebase for testing

Environment Variables

The following environment variables can be used to configure the application:

See .env.example for a template.

Development

Running Tests

# Run all tests
npm test

# Run tests with coverage
npm test -- --coverage

# Test MCP server setup
npm run test:setup

Future Improvements

• Support for more file types
• Support for alternative LLM providers
• Integration with an Electron app for a GUI interface
• Enhanced MCP server capabilities
• Advanced token usage tracking
• OpenTelemetry-based observability
• Enhanced audit logging capabilities
• Secret scanning integration

Requirements

• Node.js 18+

Installation

1. Clone the repository

   Copy

   git clone https://github.com/nicobailon/code-summarizer.git
   cd code-summarizer

2. Install dependencies:

   Copy

   npm install

3. Create a .env file with your Google API key:

   Copy

   GOOGLE_API_KEY=your_api_key_here

4. Build the project:

   Copy

   npm run build

MCP Server Setup and Integration

The code summarizer includes a Model Context Protocol (MCP) server that allows LLM tools like Claude Desktop, Cursor AI, and Cline to access code summaries and file content.

Starting the MCP Server

# Start the MCP server
npm start -- server

By default, the server runs on port 24312. You can change this in your configuration:

# Set custom MCP server port
npm start -- config set --port 8080

Connecting with Claude Desktop

1. Start the code-summarizer MCP server
2. Open Claude Desktop and click on the Claude menu, then "Settings..."
3. Navigate to the "Developer" section
4. Create a file at ~/.claude/claude_desktop_config.json (macOS/Linux) or %USERPROFILE%\.claude\claude_desktop_config.json (Windows) with this content:

{
  "code-summarizer": {
    "command": "npx",
    "args": ["-y", "your-path-to-code-summarizer/bin/code-summarizer.js", "server"],
    "env": {
      "GOOGLE_API_KEY": "your_api_key_here"
    }
  }
}

5. Restart Claude Desktop
6. After restarting, you can ask Claude to access your codebase, e.g., "Summarize the files in my project"

Example prompts for Claude Desktop:

• "Can you summarize all the JavaScript files in my project?"
• "Please give me a high-level overview of my codebase."
• "Explain what the file 'src/config/config.ts' does."
• "Find all functions related to authentication in my code."

Connecting with Cursor AI

1. Start the code-summarizer MCP server
2. Create a .cursor/mcp.json file in your project directory:

{
  "mcpServers": {
    "code-summarizer": {
      "transport": "sse",
      "url": "http://localhost:24312/sse",
      "headers": {
        "x-api-key": "your_api_key_here"
      }
    }
  }
}

3. Restart Cursor or reload your project
4. Ask Cursor about your code, e.g., "Can you summarize my codebase?"

Example prompts for Cursor:

• "Summarize the structure of this codebase for me."
• "What are the key components in this project?"
• "Give me a detailed explanation of the MCP server implementation."
• "Help me understand how the retry mechanism works."

Connecting with Cline

1. Start the code-summarizer MCP server
2. In Cline, you can add the MCP server with a command:

/mcp add code-summarizer http://localhost:24312/sse

3. Then authenticate with your API key:

/mcp config code-summarizer headers.x-api-key your_api_key_here

4. You can then ask Cline to use the code-summarizer, e.g., "Please summarize my code files"

Example prompts for Cline:

• "What does each file in my project do?"
• "Create a summary of all TypeScript files."
• "Explain the authentication flow in this codebase."
• "What are the main functions in the 'summarizer' directory?"

What You Can Do with the MCP Integration

Using the MCP integration, you can:

1. Get file summaries: Request concise explanations of what specific files do
2. Explore directories: Browse through your codebase structure
3. Batch processing: Summarize multiple files at once
4. Targeted queries: Find specific patterns or functionality in your code
5. Customize summaries: Control detail level and summary length
6. Update settings: Change configuration options through the MCP interface

The MCP server exposes your codebase to the LLM tools in a structured way, allowing them to read, navigate, and summarize your code without having to paste code snippets manually.

Usage

Command Line Interface

# Default command (summarize)
npm start -- summarize [directory] [output-file] [options]

# Summarize code in the current directory (output to summaries.txt)
npm start -- summarize

# Summarize code with specific detail level and max length
npm start -- summarize --detail high --max-length 1000

# Show help
npm start -- --help

Configuration Management

# Set your API key
npm start -- config set --api-key "your-api-key" 

# Set default detail level and max length
npm start -- config set --detail-level high --max-length 1000

# Set MCP server port (default: 24312)
npm start -- config set --port 8080

# Show current configuration
npm start -- config show

# Reset configuration to defaults
npm start -- config reset

API Authentication

When connecting to the MCP server, you need to include your API key in the request headers:

x-api-key: your_api_key_here

All endpoints (except /health) require authentication.

Troubleshooting

Common MCP Connection Issues

1. Connection Refused

   • Make sure the MCP server is running (npm start -- server)
   • Verify the port is correct in your configuration
   • Check for firewall issues blocking the connection

2. Authentication Errors

   • Verify you've added the correct API key in the headers (x-api-key)
   • Check that your API key is valid and properly formatted
   • Make sure environment variables are set correctly

3. Transport Errors

   • Ensure the correct transport type is specified (SSE)
   • Check that the URL includes the correct endpoint (/sse)
   • Verify network connectivity between the client and server

4. Permission Issues

   • Ensure the MCP server has read access to your codebase
   • Check file permissions if summarizing fails for specific files

5. Claude Desktop Not Finding the MCP Server

   • Verify the path in claude_desktop_config.json is correct
   • Make sure the command and args point to the right location
   • Check Claude Desktop logs for any configuration errors

6. Rate Limiting

   • If you see "Too many requests" errors, wait and try again later
   • Consider adjusting the rate limiting settings in the server code

For other issues, check the server logs or open an issue on the GitHub repository.