dev_buddy_mcp

MCP (Model Context Protocol) server for DevBuddy. Connects your running Flutter app's diagnostics to AI IDEs like Claude Code, Cursor, and VS Code Copilot.

How It Works

Flutter App (debug)          MCP Server              AI IDE
┌──────────────────┐    ┌─────────────────┐    ┌──────────────┐
│ DevBuddyOverlay  │    │ bin/server.dart  │    │ Claude Code  │
│   ↓               │    │                 │    │ or Cursor    │
│ DevBuddyEngine   │◄──►│ HTTP bridge     │◄──►│              │
│   ↓               │HTTP│ (localhost)     │MCP │ "What's      │
│ DiagnosticServer ─┤────┤                 ├────┤  causing     │
│ (port 8585)      │    │ 9 diagnostic    │    │  the jank?"  │
└──────────────────┘    │ tools           │    └──────────────┘
                         └─────────────────┘
  1. Your Flutter app runs with DevBuddy → collects FPS, errors, network, memory, rebuild data
  2. DevBuddy starts an HTTP server on localhost:8585 (debug only, zero cost in release)
  3. The MCP server (bin/server.dart) is launched by your AI IDE
  4. AI IDE sends tool calls → MCP server → HTTP to Flutter app → returns diagnostics
  5. AI reads the diagnostics and suggests fixes in your code

Setup

Step 1: Add DevBuddy to Your Flutter App

# pubspec.yaml
dependencies:
  dev_buddy: ^0.2.0

// main.dart
import 'package:dev_buddy/dev_buddy.dart';

MaterialApp(
  builder: (context, child) => DevBuddyOverlayImpl(
    enabled: kDebugMode,
    enableMcpServer: true,    // Starts HTTP server on localhost:8585
    mcpServerPort: 8585,      // Default port
    modules: [
      PerformanceModule(),
      ErrorTranslatorModule(),
      NetworkModule(),
      MemoryModule(),
      RebuildTrackerModule(),
    ],
    child: child!,
  ),
)

Step 2: Configure Claude Code

Add to your project's .mcp.json:

{
  "mcpServers": {
    "dev-buddy": {
      "command": "dart",
      "args": ["run", "dev_buddy_mcp"],
      "cwd": "/path/to/your/flutter/project"
    }
  }
}

Or if you installed the package globally:

{
  "mcpServers": {
    "dev-buddy": {
      "command": "dart",
      "args": ["pub", "global", "run", "dev_buddy_mcp"]
    }
  }
}

Step 3: Run Your App and Ask Claude

  1. Run your Flutter app in debug mode: flutter run
  2. You should see: 🔧 DevBuddy diagnostic server running on http://localhost:8585
  3. Open Claude Code and ask:
"Check my running Flutter app for performance issues"
"What network requests are slow?"
"Are there any memory leaks?"
"Why is the FPS dropping?"

Claude will use the MCP tools to query your app's live diagnostics and suggest fixes.

Available Tools

Tool What It Does
dev_buddy/diagnostics Compact snapshot: severity, event count, top 3 issues
dev_buddy/suggest AI-friendly fix suggestions for active problems
dev_buddy/search_events Filter events by module, severity, or text query
dev_buddy/search_network Query HTTP traffic by URL, status, duration
dev_buddy/search_state Time-travel through Riverpod/BLoC state changes
dev_buddy/detail Full event details by index
dev_buddy/performance FPS, frame timing, jank analysis
dev_buddy/memory RSS memory, growth rate, leak detection
dev_buddy/errors Translated Flutter errors with fix suggestions

Custom Port

If port 8585 is taken, use a different port:

// Flutter app
DevBuddyOverlayImpl(
  mcpServerPort: 9090,
  // ...
)

// .mcp.json
{
  "mcpServers": {
    "dev-buddy": {
      "command": "dart",
      "args": ["run", "dev_buddy_mcp", "9090"]
    }
  }
}

VS Code Extension (Copilot/Continue)

For VS Code with MCP-compatible extensions, add to .vscode/mcp.json:

{
  "servers": {
    "dev-buddy": {
      "command": "dart",
      "args": ["run", "dev_buddy_mcp"],
      "cwd": "${workspaceFolder}"
    }
  }
}

Cursor IDE

Add to Cursor's MCP settings (Settings > MCP Servers):

{
  "dev-buddy": {
    "command": "dart",
    "args": ["run", "dev_buddy_mcp"]
  }
}

Verifying the Connection

You can test the HTTP bridge directly:

# Check if Flutter app's diagnostic server is running
curl http://localhost:8585/health

# Get a diagnostic snapshot
curl http://localhost:8585/snapshot

# Call a tool directly
curl -X POST http://localhost:8585/tool \
  -H "Content-Type: application/json" \
  -d '{"tool": "dev_buddy/diagnostics", "params": {}}'

Security

  • The diagnostic server only binds to localhost (127.0.0.1) — not accessible from network
  • Only runs in debug mode — zero code in release builds
  • All data is PII-sanitized before reaching AI (emails, tokens, keys are redacted)
  • No secrets or user data leave the machine

Troubleshooting

Problem Solution
"Flutter app not reachable" Make sure flutter run is active and you see the 🔧 log message
Port already in use Change mcpServerPort in your app and update .mcp.json
No events showing Interact with your app — scroll, tap, navigate to generate diagnostics
MCP server not found Ensure dev_buddy_mcp is in your dependencies or installed globally

Libraries

dev_buddy_mcp
DevBuddy MCP — AI IDE integration via Model Context Protocol.