Coding Thinker MCP

AI Code Surgeon: Think → Edit → Verify → Commit
The only MCP that combines reasoning, surgical diffs, syntax validation, and git commits in one workflow.

🎯 What This Does

Traditional AI code editing: Copy → Paste → Format → Lint → Git commit → 10 minutes wasted.

mcp-coding-thinker: "Fix dark mode" → 90 seconds → ✅ Committed with perfect style.

The Magic Workflow

1. Analyze: Learns your code style (quotes, indentation, patterns)
2. Think: Records reasoning steps for transparency
3. Plan: Creates surgical line-by-line diffs with fuzzy matching
4. Verify: Validates syntax, checks for conflicts, generates lintable preview
5. Execute: Applies changes, auto-formats with Prettier, backs up originals, commits to git

Why It's Different

Most AI code tools either just suggest changes OR execute them. This does both with full reasoning transparency:

• Code-aware thinking: Records WHY each change is made, not just WHAT
• Surgical precision: Line-by-line diffs with overlap handling, not full-file replacements
• Zero manual work: Auto-formats + syntax checks + git commits in one flow
• Always safe: Backups every file, validates before writing, handles git edge cases
• Style-aware: Learns your existing code patterns and enforces them automatically

🚀 Installation

Claude Desktop (One Command)

claude mcp add coding-thinker -- npx -y mcp-coding-thinker

Google Gemini (One Command)

gemini mcp add coding-thinker npx -y mcp-coding-thinker

🔧 Manual Setup for Claude Desktop

macOS

Edit ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "coding-thinker": {
      "command": "npx",
      "args": ["-y", "mcp-coding-thinker"]
    }
  }
}

Windows

Edit %APPDATA%\Claude\claude_desktop_config.json:

{
  "mcpServers": {
    "coding-thinker": {
      "command": "npx",
      "args": ["-y", "mcp-coding-thinker"]
    }
  }
}

Linux

Edit ~/.config/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "coding-thinker": {
      "command": "npx",
      "args": ["-y", "mcp-coding-thinker"]
    }
  }
}

Restart Claude Desktop after saving.

🔧 Manual Setup for Google Gemini

For Google AI Studio Desktop (if supported)

Create/edit MCP config file (location varies by OS):

• macOS/Linux: ~/.config/gemini/mcp_config.json
• Windows: %APPDATA%\Gemini\mcp_config.json

{
  "mcpServers": {
    "coding-thinker": {
      "command": "npx",
      "args": ["-y", "mcp-coding-thinker"]
    }
  }
}

🔧 Using with Any MCP-Compatible Client

# Start the server with npx
npx -y mcp-coding-thinker

# Or test with MCP Inspector
npx @modelcontextprotocol/inspector npx -y mcp-coding-thinker

📖 Usage Example

Prompt to AI (Claude/Gemini):

Use coding-thinker to fix the dark mode color contrast in Header.tsx

What Happens:

1. Analyzes Header.tsx → Detects React + Tailwind + 2-space indent
2. Records thoughts: "Dark text on dark bg detected at line 42"
3. Creates diff: Change text-gray-900 → text-gray-100
4. Validates syntax ✅
5. Formats with Prettier ✅
6. Backs up to .mcp-backups/ ✅
7. Commits: "mcp-coding-thinker: Fix dark mode contrast" ✅

Result: Production-ready commit in 90 seconds.

🛠️ How It Works

4 Tools

1. analyze_context

• Input: Files + request
• Output: Session ID, style profile, issues, research queries
• Use: Always call first

2. think_aloud

• Input: Session ID + reasoning step
• Output: Recorded thought + depth score
• Use: Show incremental reasoning (5+ steps = excellent quality)

3. plan_and_verify

• Input: Session ID + diffs (line-by-line changes)
• Output: Validation results, safety score, lintable preview
• Use: Submit exact diffs for verification

4. execute_and_review

• Input: Session ID + approval
• Output: Execution status, git operations, mirror critique prompt
• Use: Apply changes with full safety net

🔒 Safety Features

Git Protection

• ✅ Detects detached HEAD, rebase/merge in progress
• ✅ Handles pre-commit hook failures gracefully
• ✅ Provides revert instructions: git revert HEAD

Backup System

• ✅ Copies originals to .mcp-backups/{session-id}/
• ✅ Preserves directory structure
• ✅ 1-hour session TTL (auto-cleanup)

Validation

• ✅ Syntax check before formatting (no silent fails)
• ✅ Fuzzy diff matching (±2 lines tolerance)
• ✅ Overlapping diff detection with unified patching
• ✅ 80% similarity threshold for stale diffs
• ✅ Size limits on diffs (prevents memory issues)
• ✅ Cross-platform line ending normalization (CRLF/LF)
• ✅ Prettier config detection from user's repo

🎓 Best Practices

For AI Models (Claude/Gemini)

1. Always analyze first: Call analyze_context before planning
2. Think deeply: 5+ think_aloud calls = 95%+ confidence
3. Exact diffs: Provide real line numbers from original files
4. Dry run first: Use dry_run=true to preview before executing

For Users

• Commit changes before using (or work on feature branch)
• Review mirror critique prompt for architectural insights
• Use research queries to verify best practices
• Check .mcp-backups/ if rollback needed

🔧 Development

git clone https://github.com/awssat/mcp-coding-thinker.git
cd mcp-coding-thinker
npm install
npm run build

# Test with MCP Inspector
npm run inspector

🤝 Contributing

PRs welcome! Priority areas:

• Additional language support
• Test coverage
• Performance optimizations

📜 License

MIT © 2026