Govinda-Fichtner/debugger-mcp

DAP MCP Server

Enable AI agents to programmatically debug applications across multiple languages through a unified MCP interface.

What is This?

A Rust-based MCP (Model Context Protocol) server that exposes debugging capabilities to AI assistants (Claude, Gemini CLI, etc.) by bridging to the Debug Adapter Protocol (DAP).

In short: AI agents can set breakpoints, step through code, inspect variables, and investigate bugs autonomously across Python, Ruby, Node.js, Go, and Rust.

Quick Start

1. Install

Docker (Recommended):

# Choose image for your language
docker build -f Dockerfile.python -t debugger-mcp:python .
docker run -i debugger-mcp:python

Native:

cargo build --release
./target/release/debugger_mcp serve

2. Configure Claude Desktop

{
  "mcpServers": {
    "debugger": {
      "command": "/path/to/debugger_mcp",
      "args": ["serve"]
    }
  }
}

3. Debug

Start debugging from Claude!

Detailed setup:

Status

🎉 Production-Ready - Multi-Language Support

Supported Languages: Python, Ruby, Node.js, Go, Rust (all 100% functional)

Continuous Integration

Workflow	Purpose	Latest Status
CI	Code quality, security, unit tests (193 tests)
Integration Tests	End-to-end AI debugging: 5 languages × 2 AI clients (Claude Code + Codex)

What Integration Tests Do: Real AI agents (Claude Code and Codex) autonomously debug programs via MCP, validating 8 debugging operations per test across all languages.

Latest Results (10/10 tests passing):

Legend: S=Session Start, B=Breakpoint, C=Continue, T=Trace, E=Evaluate, D=Disconnect

Understanding CI: See

Features

Supported Languages

Debugging Capabilities

✅ Current:

• Start/stop debugging sessions
• Set source breakpoints
• Execution control (continue, step over/into/out, pause)
• Expression evaluation
• Stack trace inspection
• Variable inspection

⏳ Planned:

• Conditional breakpoints & logpoints
• Exception breakpoints
• Multi-threaded debugging
• Remote debugging
• Data breakpoints

Architecture

AI Agent (Claude, Gemini, etc.)
    ↕ MCP Protocol (JSON-RPC)
┌──────────────────────────────────┐
│   DAP MCP Server (Rust/Tokio)   │
│ ┌─────────────────────────────┐  │
│ │  MCP Layer (Tools/Resources)│  │
│ └──────────┬──────────────────┘  │
│ ┌──────────┴──────────────────┐  │
│ │  Language-Agnostic Core     │  │
│ └──────────┬──────────────────┘  │
│ ┌──────────┴──────────────────┐  │
│ │  DAP Protocol Client        │  │
│ └─────────────────────────────┘  │
└──────────┼───────────────────────┘
           ↕ Debug Adapter Protocol
    ┌──────┴──────┐
debugpy  rdbg  delve  CodeLLDB
(Python)(Ruby) (Go)  (Rust/C++)

Deep dive:

Usage Example

User: "My Python script crashes. Can you debug it?"

Claude:
  → debugger_start(language="python", program="/workspace/script.py")
  → debugger_set_breakpoint(sourcePath="/workspace/script.py", line=42)
  → debugger_continue()
  → debugger_wait_for_stop()
  [Program stops at breakpoint]
  → stack = debugger_stack_trace()
  → debugger_evaluate(expression="user_data")

  "The crash occurs because 'user_data' is None when fetch_user() fails.
   The code doesn't check for None before accessing user_data.name..."

Expression syntax by language:

Common Issues

❓ Breakpoint not verified? → Ensure debug symbols: -g flag for rustc/gcc, debugpy for Python → Check source path matches exactly

❓ Session timeout? → Verify debugger installed: pip install debugpy, gem install debug, etc. → Check debugger in PATH

❓ Docker path issues? → Use container paths: /workspace/... (not host paths like /home/user/...) → Ensure volume mounted correctly

Full guide:

Documentation

By Use Case

🚀 Getting started? →

🐳 Deploying with Docker? →

🏗️ Understanding architecture? →

➕ Adding a new language? →

✅ Understanding CI/CD? →

🐛 Troubleshooting issues? →

🧪 Writing tests? →

Documentation Structure

• - System design, components, technical decisions
• - Developer guides, testing, setup
• - Deployment, Docker, expressions, troubleshooting
• - CI/CD, releases, cross-platform builds

Complete index:

Development

Prerequisites

• Rust 1.70+ (rustup update)
• Docker (for integration tests)
• Language-specific debuggers (for testing):
  • Python: pip install debugpy
  • Ruby: gem install debug
  • Node.js: npm install -g node-debug2

Build & Test

# Clone
git clone https://github.com/Govinda-Fichtner/debugger-mcp.git
cd debugger-mcp

# Install pre-commit hooks (recommended)
pre-commit install --install-hooks
pre-commit install --hook-type commit-msg
pre-commit install --hook-type pre-push

# Build
cargo build --release

# Run unit tests
cargo test

# Run integration tests (requires debuggers)
cargo test --test '*integration*' -- --ignored

Pre-commit Hooks

Automated quality checks run before commit/push:

• Formatting (cargo fmt)
• Linting (cargo clippy)
• Unit tests
• Security scanning (gitleaks, cargo-audit)
• Code coverage (60% minimum)

Setup:

Contributing

We welcome contributions! See for:

• Development setup
• Architecture overview
• Testing guidelines
• Code style

Contribution workflow:

1. Fork repository
2. Create feature branch
3. Make changes with tests
4. Run pre-commit run --all-files
5. Submit pull request

Roadmap

✅ Completed Phases

• Phase 0: Research & Architecture
• Phase 1: MVP - Python Support
• Phase 2: Ruby Validation
• Phase 3: Multi-Language Support (Python, Ruby, Node.js, Go, Rust)

🚧 Current Phase

Phase 4: Production Features

• Conditional breakpoints
• Exception handling
• Security hardening
• Performance optimization

📅 Future Phases

Phase 5: Community

• Open source release
• Plugin API
• VS Code extension
• Additional languages (Java, C#, PHP)

Technology Stack

License

TBD (likely MIT or Apache 2.0)

Built with ❤️ and 🦀 using Rust

Last Updated: 2025-10-19