Impart MCP

🚀 Impart MCP (Agent Orchestration Layer)

impart-mcp is an agent orchestration server based on the Model Context Protocol (MCP). It serves as the "Agent Orchestration Layer", encapsulating expert agents with high engineering capabilities (such as Claude Code, Codex CLI, etc.) into standard MCP tools, enabling host environments (Claude Desktop, Cursor, etc.) to call these "super agents" capable of handling complex tasks through a unified interface.

🚀 Quick Start

impart-mcp is an intelligent agent orchestration server based on the Model Context Protocol (MCP). It acts as the "Agent Orchestration Layer", encapsulating expert agents with high engineering capabilities (such as Claude Code, Codex CLI, etc.) into standard MCP tools, enabling host environments (Claude Desktop, Cursor, etc.) to call these "super agents" capable of handling complex tasks through a unified interface.

✨ Features

• Multi-agent Orchestration: Comes with 5 professional agents with preset expert prompts and tool permissions.
• Cross-platform Integration: Perfectly compatible with MCP-supported environments such as Claude Desktop, Cursor, and Gemini CLI.
• Asynchronous Task Processing: Supports background execution of time-consuming tasks and result polling.
• Mixed Model Providers: Integrates the Vercel AI SDK, supporting agents powered by Claude Code, Codex CLI, and Gemini CLI.
• Flexible Configuration: Customize the model, switch, and parameters of each agent through a JSON configuration file.

🛠️ Expert Agents

📦 Installation

Method 1: Run Directly with npx (Recommended)

No installation required. Use it directly in the MCP configuration:

npx impart-mcp@latest

Method 2: Global Installation

npm install -g impart-mcp
# or
bun add -g impart-mcp

Method 3: Install from Source Code

Prerequisites: Bun or Node.js >= 18

git clone <repository-url>
cd impart-mcp
bun install  # or npm install
bun run build  # Build the product

🔌 Integration

Configure in Cursor

1. Open Cursor settings -> Features -> MCP.
2. Add a new MCP Server.
3. Name: impart-mcp
4. Type: command
5. Command: npx impart-mcp@latest (or use impart-mcp directly after global installation)

Configure in Claude Code

Use the claude mcp add command to add it in one step:

claude mcp add impart-mcp -- npx impart-mcp@latest

Or after global installation:

claude mcp add impart-mcp -- impart-mcp

⚙️ Configuration

By default, all agents can run with the built-in default configuration without additional configuration.

Custom Configuration (Optional)

If you need to customize the model or switch of an agent, you can specify the configuration file path through the environment variable AGENT_CONFIG_PATH:

{
  "mcpServers": {
    "impart-mcp": {
      "command": "npx",
      "args": ["impart-mcp@latest"],
      "env": {
        "AGENT_CONFIG_PATH": "/path/to/your/agent-config.json"
      }
    }
  }
}

Automatic Initialization: If the specified configuration file does not exist, the system will automatically generate a configuration template containing all default values during the first run, making it easy for you to modify as needed.

Configuration File Example (agent-config.json):

{
  "advisor": {
    "model": "codex/gpt-5.2",
    "enabled": true
  },
  "researcher": {
    "model": "claude/sonnet",
    "enabled": true
  },
  "frontend-ui-ux-engineer": {
    "model": "gemini/gemini-3-pro-preview",
    "enabled": true
  }
}

📋 API Documentation (MCP Tools)

The server provides the following core tools:

Parameter Explanation

call_agent / call_agent_async:

• agent: Agent name enumeration (advisor | researcher | explore | frontend-ui-ux-engineer | document-writer)
• prompt: Task description
• cwd: Working directory (required)
• context: Optional context information

👨‍💻 Development

Core Logic Paths

📄 License

MIT License