Vibecode MCP Server

A Node.js-based Model Context Protocol (MCP) server that exposes your local .agent and .gemini rules and skills to AI clients.

Features

• Local Context: Reads .gemini/GEMINI.md and .agent/memory/ARCHITECTURE.md as MCP Resources.
• Skills as Prompts: Scans .agent/skills/ and provides them as MCP Prompts, enabling quick execution of workflows.
• Dual Transports:
  • Standard I/O mode for local IDE bindings (e.g., Cursor, Claude Desktop).
  • HTTP Server-Sent Events (SSE) mode protected by an API Key for remote bindings.

Installation

npm install -g vibecode-mcp

Usage

Local Usage (Stdio Mode)

Run the server via standard I/O (often used by Claude Desktop):

vibecode-mcp start-stdio --root /path/to/your/project

(If --root is omitted, it defaults to the current working directory).

Public Usage (SSE Mode with API Key)

Run the server opening an HTTP port, accessible over the network but protected by a Bearer Token:

vibecode-mcp start-sse --port 3000 --key YOUR_SUPER_SECRET_KEY --root /path/to/your/project

An AI agent/client can then connect to: http://<SERVER_IP>:3000/sse

Providing the exact following header: Authorization: Bearer YOUR_SUPER_SECRET_KEY

Architecture

• vibecode://context/global-rules -> Evaluates to <root>/.gemini/GEMINI.md.
• vibecode://context/architecture -> Evaluates to <root>/.agent/memory/ARCHITECTURE.md.
• Prompts -> Dynamically lists skills located under <root>/.agent/skills/.

Roadmap

• Add support for invoking .gemini/commands via MCP Tools.