🚀 MCP Commander: Token-Efficient AI Platform

A modular Model Context Protocol (MCP) Server focused on persistent memory, project tracking, and dashboard-driven operations for local AI agent workflows.

✨ Core Philosophy

Stop Amnesia, Stop Unnecessary Iterations. This MCP Server empowers your AI Agent with a permanent memory architecture, rigorous context constraints, and a dynamic plugin ecosystem.

🌟 Key Features

🧩 1. Dynamic Plugin Matrix (Namespace Architecture)

Tools are grouped into plugins and loaded through the MCP server at startup.

• Plugin Metadata: Each plugin is described by plugin.yaml, including its tools, owner, and management surface.
• Core Plugins Included:
  • core_system: File manipulators, terminal execution, codebase analysis.
  • core_zero_waste: Context guards, AI Sandboxing (Patch diffing), semantic Vector Memory query.
  • github_integration: Native multi-stage Git & GitHub PR pipelines.
  • antigravity_sync: Direct memory tunneling to native AI UI Artifacts (Walkthroughs & Plans).

🖥️ 2. MCP Commander Dashboard (Streamlit & Docker)

A Streamlit dashboard for tracking projects, sprints, backlog items, logs, and plugin metadata.

• Agile Backend: Manage Projects, Sprints, and Backlogs directly inside PostgreSQL.
• Tool Footprints: Track tool invocations with sanitized parameters and summarized responses in system_tool_logs.
• Toggle Features: Hot-swap/Enable/Disable AI Plugins via a GUI directly writing to plugin.yaml.
• Antigravity Brain Logs: Visualize historical Architectures and Walkthroughs generated by previous AI sessions.

🧠 3. Persistent Semantic Memory (pgvector)

Built-in PostgreSQL + pgvector instances containerized by Docker. Semantic retrieval is available when the embedding model is installed and reachable in the target environment.

---

⚡ Quickstart

1. Requirements

• Docker & Docker Compose
• Python 3.10+ (If running outside Docker)

2. Environment Setup

Configure your environment mapping. In the root directory or mcp_server/.env:

# mcp_server/.env
DATABASE_URL=postgresql://mcp:mcp@db:5432/mcp_server_db
VECTOR_BACKEND=pgvector

# Mapping your Native AI Workspace generated artifacts to the Dashboard:
HOST_BRAIN_DIR=C:\Users\YOUR_NAME\.gemini\antigravity\brain

3. Launch

The default stack (DB, Dashboard, MCP Server) boots in a single command:

docker compose up -d --build

4. Endpoints

• Streamlit Web Dashboard: http://localhost:8501
• MCP Server (SSE Endpoint): http://localhost:8000/mcp/sse

💻 VS Code / Antigravity Editor Setup

1. Open the workspace folder in VS Code: d:\WORK\1.MCP_SERVER.
2. Install recommended extensions: Python, Pylance, Docker, and Live Share (optional).
3. Create .vscode/launch.json using the following snippet:

{
  "version": "0.2.0",
  "configurations": [
    {
      "name": "Python: Run MCP server",
      "type": "python",
      "request": "launch",
      "program": "${workspaceFolder}/mcp_server/main.py",
      "console": "integratedTerminal",
      "args": ["--sse"]
    }
  ]
}

4. Ensure mcp_server/.env is configured and mcp_server/requirements.txt installed: pip install -r mcp_server/requirements.txt.

Antigravity Editor (Gemini UX) setup

1. Confirm mapping in .env:

HOST_BRAIN_DIR=C:\Users\YOUR_NAME\.gemini\antigravity\brain

2. Start the Antigravity local UI (if using local Gem workspace) and point its project root to d:\WORK\1.MCP_SERVER.
3. Open readme.md, plugin scripts and mcp_server/main.py in the Antigravity editor.
4. Start MCP server (WS and SSE) from terminal:

cd d:\WORK\1.MCP_SERVER
docker compose up -d --build

5. Use the Antigravity code browser to inspect plugins/* and in-editor tool execution logs.

⚠️ Current Scope

• The project is under active remediation and should be treated as development-stage, not production-ready.
• Git tools require the Docker workspace mount and git binary provided by the bundled mcp_server/Dockerfile.
• Semantic memory depends on the configured embedding model and vector backend availability.
• Authentication and fine-grained access control are not implemented yet.

---

🏗️ Architecture

graph TD
    AI[AI Agent Client] -->|SSE/JSON-RPC| MCP(MCP Server Core)
    MCP --> Router[Plugin Router Auto-Loader]
    Router --> P1(Core Zero Waste)
    Router --> P2(Github Integration)
    Router --> P3(...)
    MCP <--> DB[(PostgreSQL + pgvector)]
    Streamlit[Web Dashboard] <--> DB
    Streamlit <--> Yaml(Plugin Yaml Configs)

Loading

🛡️ License

MIT License

⚖️ Disclaimer

• MCP Commander is provided "as-is".
• Users are fully responsible for deployment, operations, and security in their environments.
• No warranty for performance, security, or uptime. Intended for testing and development.
• Avoid using directly in production until auth/RBAC, logging, and error controls are fully implemented.