🧠 Thinker MCP

Think before you code. Thinker MCP is a Model Context Protocol server that gives AI coding assistants persistent project memory and structured reasoning.

Every prompt passes through 3 phases before the AI writes a single line of code:

Clarify → Context → Chain Think

✨ What It Does

| Problem | Thinker's Solution | |---|---| | AI forgets everything between sessions | Auto-maintained PROJECT_MAP + CHANGE_LOG | | Vague prompts → vague code | Ambiguity detection + clarifying questions | | Decisions get reversed accidentally | Decision memory across all sessions | | Changes break other files silently | Pre-change impact analysis | | No reasoning before coding | Forced step-by-step chain-of-thought |

🚀 Quick Start

Install (zero setup)

npx -y thinker-mcp

Configure Your IDE

Kiro (AWS) — .kiro/mcp.json:

{
  "mcpServers": {
    "thinker-mcp": {
      "command": "npx",
      "args": ["-y", "thinker-mcp"],
      "env": {
        "MCP_MODE": "balanced"
      }
    }
  }
}

Antigravity (Google) — .antigravity/mcp.json:

{
  "servers": [
    {
      "name": "thinker-mcp",
      "transport": "stdio",
      "command": "npx",
      "args": ["-y", "thinker-mcp"]
    }
  ]
}

First Use

1. Open your project in your IDE
2. Run: initialize_project tool with your project root path
3. Thinker scans all files, creates .mcp-context/ folder
4. Start coding! Every prompt now flows through Clarify → Context → Chain Think

🛠️ The 7 Tools

| Tool | Phase | What It Does | |---|---|---| | initialize_project | Setup | Scans project, creates context docs | | clarify_prompt | Phase 1 | Detects ambiguity, asks targeted questions | | load_context | Phase 2 | Loads relevant project history & decisions | | chain_solve | Phase 3 | Forces step-by-step reasoning before coding | | pre_change_impact | Safety | Analyzes what could break before changes | | update_docs | Post | Auto-updates PROJECT_MAP + CHANGE_LOG | | health_check | Maintenance | Verifies docs are fresh and in sync |

🎯 User Modes

| Mode | Who It's For | Behavior | |---|---|---| | 🟢 GUIDE | Beginners | Max questions, plain English, all warnings | | 🔵 BALANCED | Solo devs | Smart clarification, skip if clear | | 🔴 EXPERT | Seniors | Skip unless critical, deep tradeoff analysis | | 🟡 AUDIT | Teams | Track who prompted what, conflict detection |

📁 Context Documents

Thinker creates a .mcp-context/ folder in your project with:

• PROJECT_MAP.md — What every file does, its exports, dependencies, and risk level
• CHANGE_LOG.md — Every prompt, every decision, every file change (append-only)
• SESSION_STATE.json — Current mode, prompt count, known answers
• DECISIONS.md — Decision tracking (v1.1)
• DANGER_ZONES.md — High-risk file tracking (v1.1)
• TEAM_CONVENTIONS.md — Team coding conventions (v1.1)

🔧 Development

# Install dependencies
npm install

# Run in development
npm run dev

# Build for production
npm run build

# Start production server
npm start

📦 Publishing

npm run build
npm login
npm publish --access public

After publishing: npx -y thinker-mcp — zero setup, no global install.

License

MIT