multi-llm-cross-check

This MCP server provides a cross-check capability by querying multiple large language model providers in parallel. When configured, Claude Desktop can communicate with this MCP server to send a prompt and receive responses from the configured providers such as OpenAI (ChatGPT), Anthropic (Claude), Perplexity AI, and Google Gemini. The server handles parallel requests, aggregates responses, and returns a structured dictionary mapping each provider to its respective output. To use it, enable the MCP server in Claude Desktop via the provided configuration, then invoke the cross_check tool in your conversation and supply a prompt. The server will return an object with individual results for each enabled provider, or skip a provider if its API key is not configured.

Prerequisites:

• Python 3.8 or higher
• uv package manager (pip install uv)
• API keys for the LLM providers you plan to use

Installation via Smithery (automatic):

1. Install the MCP server with Smithery:

npx -y @smithery/cli install @lior-ps/multi-llm-cross-check-mcp-server --client claude

Manual installation:

1. Clone the repository:

git clone https://github.com/lior-ps/multi-llm-cross-check-mcp-server.git
cd multi-llm-cross-check-mcp-server

2. Initialize a uv environment and install requirements:

uv venv
uv pip install -r requirements.txt

3. Configure Claude Desktop to connect to the MCP server (example shown in the repo README):

• Create claude_desktop_config.json in Claude Desktop configuration with the appropriate mcp_servers entry:

{
  "mcp_servers": [
    {
      "command": "uv",
      "args": [
        "--directory",
        "/multi-llm-cross-check-mcp-server",
        "run",
        "main.py"
      ],
      "env": {
        "OPENAI_API_KEY": "your_openai_key",
        "ANTHROPIC_API_KEY": "your_anthropic_key",
        "PERPLEXITY_API_KEY": "your_perplexity_key",
        "GEMINI_API_KEY": "your_gemini_key"
      }
    }
  ]
}

Notes:

1. Enable only the providers for which you have API keys; missing keys will cause that provider to be skipped.
2. You may need to specify the full path to the uv executable in the command field if necessary.

Tips:

• Ensure all API keys are kept secure and not exposed in public config files.
• If a provider is unavailable or returns an error, the MCP server will still return results from the other providers.
• The system is designed for asynchronous parallel processing; prompts should be crafted with the understanding that response times may vary by provider.
• You can adjust which providers are included by editing the environment variables in the mcp_config entry or the Claude Desktop configuration.