MCP Codex

🚀 MCP-Codex Project Documentation

MCP-Codex is a service designed to coordinate Model Context Protocol (MCP) servers. It enables AI assistants to dynamically discover and utilize tools without restarting the service. At its core, there is an intelligent agent called "The Librarian", which is responsible for understanding user needs and connecting them with appropriate tools.

📋 Problem Statement

Traditional MCP implementations have the following limitations:

1. Static Configuration: Adding new tools requires restarting the AI service.
2. Local Execution: All tools must run locally.
3. Limited Discovery Mechanism: The tool discovery and selection process lacks flexibility.
4. High Cognitive Load: Users need to manually manage tools.

💡 Solution

MCP-Codex offers the following solutions:

1. Dynamic Tool Discovery: AI assistants can automatically discover and use new tools as needed.
2. Standardized Interface: A unified interface for tool discovery and execution.
3. Remote Execution Support: Tools can run anywhere.
4. Centralized Management: Provides single - point control for tool deployment and updates.

📐 Project Architecture

Core Components

1. Codex Service

   • Provides a REST API for tool search and invocation.
   • Responsible for tool registration, discovery, and execution.

2. The Librarian

   • An intelligent agent responsible for complex tool selection logic.
   • Maintains an industry knowledge base and an internal document database.

3. Tool Repository

   • Stores all available tools and their metadata.
   • Supports tool version control and compatibility checks.

💻 Usage Examples

Start the Codex Service

python codex-serv.py

Start the MCP Client

python mcp-codex.py

Integrate with Claude Desktop

1. Install the MCP client:

   mcp install mcp-codex.py
   

2. Register two tools:
   • search: Find relevant tools.
   • call_tool: Execute the discovered tools.

🗺️ Roadmap

The Librarian (Detailed Version)

• As a Proxy RAG System: Combines retrieval and generative intelligence.
• Maintain Multiple Knowledge Sources:
  • An industry technical information database.
  • An internal tool documentation library.
  • A real - time MCP ecosystem index.
• Execute Complex Tool Selection Logic:
  • Analyze user queries to identify intentions.
  • Match the required functions.
  • Find the optimal tools across the entire MCP ecosystem.
  • Consider tool compatibility, performance, and reliability.

📚 Documentation

Tool Registration Configuration

Tools are defined in the config/codex.yaml file with the following structure:

tools:
  - name: tool_name
    description: Tool description
    server: path/to/tool/server.py
    methods:
      - method_name
    schemas:
      method_name:
        type: object
        properties:
          # Parameter definition

API Endpoints

• /health: Health check endpoint.
• /search: Search for matching tools.
• /call_tool: Execute a specific tool method.
• /reload: Reload tool configurations.

🎯 Project Significance

1. Enable True Proxy Functionality: Allows AI assistants to dynamically discover and use tools based on user needs.
2. Reduce Operational Complexity: Adds new capabilities without restarting the service.
3. Centralized Tool Management: Provides single - point control for tool deployment and updates.
4. Standardize Tool Interaction: Creates a consistent interface for tool discovery and execution.
5. Support Remote Execution: Tools can run anywhere.

📄 License

This project is licensed under the MIT License. Please refer to the LICENSE file for specific license details.