debug-mcp-server
# Debug MCP Server
An MCP (Model Context Protocol) server that bridges web browsers and Claude Code for seamless debugging workflows.
## Overview
This server receives debug data from browser widgets (console logs, error stacks, screenshots) and exposes it to Claude Code via the MCP protocol. It also facilitates interactive Q&A between Claude Code and browser users.
## Features
- **Debug Data Ingestion**: HTTP endpoints to receive debug reports from browsers
- **MCP Tools**: Expose debug data to Claude Code through MCP protocol
- **Interactive Q&A**: WebSocket-based bidirectional communication for clarification questions
- **File Storage**: Organized filesystem storage for reports and screenshots
- **Localhost Only**: Secure local-only server for development use
## Quick Start
### Installation
```bash
# Global installation
npm install -g @yourco/debug-mcp-server
# Or use npx
npx @yourco/debug-mcp-server
```
### Running the Server
```bash
# Start the server
debug-mcp-server
# Or with custom port
DEBUG_MCP_PORT=4000 debug-mcp-server
```
### Configuration
Create `~/.debug-mcp/config.json`:
```json
{
"port": 3000,
"storagePath": "~/.debug-mcp",
"retentionDays": 7
}
```
Or use environment variables:
- `DEBUG_MCP_PORT` - HTTP server port (default: 3000)
- `DEBUG_MCP_STORAGE_PATH` - Storage directory (default: `~/.debug-mcp`)
- `DEBUG_MCP_RETENTION_DAYS` - Report retention period (default: 7)
## MCP Integration
### Connect Claude Code to Server
Add to your Claude Code MCP configuration:
```json
{
"mcpServers": {
"debug-reports": {
"command": "node",
"args": ["/path/to/debug-mcp-server/dist/index.js"]
}
}
}
```
### Available MCP Tools
- `submit_debug_report` - Receive debug data from browser
- `list_debug_reports` - List all stored debug reports
- `get_debug_report` - Retrieve specific debug report by ID
- `ask_user_questions` - Send questions to browser and wait for answers
- `clear_old_reports` - Clean up old debug data
### Available MCP Resources
- `debug://reports/latest` - Get most recent debug report
- `debug://reports/{id}` - Get specific report by ID
- `debug://screenshots/{id}` - Access screenshot files
## API Endpoints
### POST /api/debug
Receive debug data from browser widget.
**Request Body:**
```json
{
"logs": [
{ "level": "error", "message": "Error message", "timestamp": 1234567890 }
],
"error": {
"message": "TypeError: Cannot read property...",
"stack": "Error stack trace..."
},
"screenshot": "data:image/png;base64,...",
"comment": "User description of the issue",
"url": "https://example.com/page",
"timestamp": 1234567890
}
```
**Response:**
```json
{
"success": true,
"data": {
"reportId": "abc123",
"timestamp": 1234567890
}
}
```
### POST /api/questions/answer
Receive answers from browser to Claude's questions.
**Request Body:**
```json
{
"sessionId": "session-123",
"answers": [
{ "questionId": "q1", "answer": "User's answer" }
]
}
```
## Development
### Setup
```bash
# Clone the repository
git clone https://github.com/yourco/debug-mcp-server.git
cd debug-mcp-server
# Install dependencies
npm install
# Run in development mode
npm run dev
```
### Project Structure
```
debug-mcp-server/
├── src/
│ ├── index.ts # Entry point
│ ├── mcp/ # MCP server implementation
│ ├── http/ # HTTP/WebSocket servers
│ ├── storage/ # File storage layer
│ ├── types/ # TypeScript types
│ └── config/ # Configuration
├── specs/ # SpecKit specifications
├── tests/ # Jest tests
└── logs/ # Session logs
```
### Testing
```bash
# Run unit tests
npm test
# Run with coverage
npm run test:coverage
# Run linter
npm run lint
```
### Building
```bash
# Build for production
npm run build
# Output: dist/index.js
```
## Architecture
```
┌─────────────────────┐
│ Browser Widget │
│ (HTTP/WebSocket) │
└──────────┬──────────┘
│
↓
┌─────────────────────────────────────┐
│ Debug MCP Server │
│ ┌─────────────────────────────┐ │
│ │ HTTP Server (Express) │ │
│ │ - POST /api/debug │ │
│ │ - POST /api/questions │ │
│ └─────────────────────────────┘ │
│ ┌─────────────────────────────┐ │
│ │ WebSocket Server │ │
│ │ - Push questions to browser│ │
│ └─────────────────────────────┘ │
│ ┌─────────────────────────────┐ │
│ │ MCP Server │ │
│ │ - Tools & Resources │ │
│ └─────────────────────────────┘ │
│ ┌─────────────────────────────┐ │
│ │ File Storage │ │
│ │ - ~/.debug-mcp/ │ │
│ └─────────────────────────────┘ │
└──────────┬──────────────────────────┘
│
↓
┌─────────────────────┐
│ Claude Code │
│ (VS Code) │
└─────────────────────┘
```
## Storage Structure
```
~/.debug-mcp/
├── reports/
│ ├── 2025-11-26/
│ │ ├── report-abc123.json
│ │ └── report-def456.json
│ └── 2025-11-27/
│ └── report-ghi789.json
├── screenshots/
│ ├── screenshot-abc123.png
│ └── screenshot-def456.png
└── config.json
```
## Troubleshooting
### Server won't start
Check if port is already in use:
```bash
lsof -i :3000
```
Use a different port:
```bash
DEBUG_MCP_PORT=4000 debug-mcp-server
```
### Claude Code can't connect
1. Verify server is running: `curl http://localhost:3000/api/health`
2. Check MCP configuration in Claude Code settings
3. Restart VS Code after config changes
### Browser widget can't send data
1. Verify server is running on localhost:3000
2. Check browser console for CORS errors
3. Ensure browser has network access to localhost
## Contributing
See [CLAUDE.md](./CLAUDE.md) for development guidelines and project constitution.
## License
MIT
## Support
For issues and feature requests, please open an issue on GitHub.
TDQS
Scored across 10 tools
Each tool targets a distinct resource and action. The two resource types (debug reports vs backlog items) are clearly separated, and the backlog action verbs (process, resolve, dismiss, update) have distinct meanings.
All tool names follow a consistent verb_noun pattern with lowercase and underscores. Verbs such as list, get, clear, process, resolve, dismiss, and update are uniformly applied.
10 tools is well within the ideal range for a focused debugging workflow. Each tool addresses a necessary operation without unnecessary bloat.
The tool set covers the full lifecycle of backlog items (pending, processing, resolved/dismissed) and provides essential report management (list, get, clear old). The inclusion of ask_user_questions fills a critical interactive gap, making the surface comprehensive.