概要
What is MCP pour OpenAI et Claude?
A simple server that acts as a Master Control Program (MCP) for interacting with AI models from OpenAI and Anthropic (Claude).
How to use MCP pour OpenAI et Claude?
Clone the repository, install dependencies with npm install, configure your API keys in a .env file based on .env.example, then start the server with npm start. The server runs on http://localhost:3000 and exposes unified and provider-specific API endpoints.
Key features of MCP pour OpenAI et Claude
- Unified API endpoint with automatic provider detection.
- Provider-specific endpoints for OpenAI and Claude.
- Supports multiple OpenAI and Claude models.
- Configurable prompt, model, maxTokens, and provider.
- Includes a simple web test interface (test-mcp.html).
Use cases of MCP pour OpenAI et Claude
- Interact with OpenAI or Claude models through a single API.
- Automatically route requests to the correct AI provider based on model name.
- Build applications that need to switch between GPT-4 and Claude models.
- Test and compare responses from different AI models easily.
FAQ from MCP pour OpenAI et Claude
What models are supported?
OpenAI models include gpt-4, gpt-4-turbo, gpt-3.5-turbo, and others. Claude models include claude-3-opus-20240229, claude-3-sonnet-20240229, claude-3-haiku-20240307, and others.
How do I configure my API keys?
Copy .env.example to .env and edit the file to add your OPENAI_API_KEY and ANTHROPIC_API_KEY values. The server reads these environment variables on startup.
What endpoints does the server provide?
Three endpoints: POST /api/mcp (unified, auto-detects provider), POST /api/mcp/openai (always OpenAI), and POST /api/mcp/claude (always Claude). All accept a JSON body with prompt, model, maxTokens, and provider parameters.
How do I choose between OpenAI and Claude?
Set the provider field to "openai", "claude", or "auto" (default) in your request. Alternatively, use the provider-specific endpoints /api/mcp/openai or /api/mcp/claude.
Is there a test interface included?
Yes, a simple web test page is included at test-mcp.html. Open it in your browser to test the MCP without writing additional code.
