Multi LLM Cross-Check MCP Server
@lior-ps
A Model Control Protocol (MCP) server that allows cross-checking responses from multiple LLM providers simultaneously
Overview
What is Multi LLM Cross-Check MCP Server?
A Model Control Protocol (MCP) server that allows cross-checking responses from multiple LLM providers simultaneously. It integrates with Claude Desktop to provide a unified interface for querying OpenAI, Anthropic, Perplexity, and Google Gemini APIs.
How to use Multi LLM Cross-Check MCP Server?
Install via Smithery (npx -y @smithery/cli install @lior-ps/multi-llm-cross-check-mcp-server --client claude) or manually by cloning the repository, setting up a uv environment, and configuring API keys in claude_desktop_config.json. Once configured, the server starts automatically with Claude Desktop; use the cross_check tool by asking to "cross check with other LLMs" and providing a prompt.
Key features of Multi LLM Cross-Check MCP Server
- Query multiple LLM providers in parallel
- Supports OpenAI, Anthropic, Perplexity, and Google Gemini
- Asynchronous parallel processing for faster responses
- Easy integration with Claude Desktop via MCP
- Independent error handling per provider
Use cases of Multi LLM Cross-Check MCP Server
- Compare answers to the same prompt across different LLMs
- Verify factual consistency by cross-checking responses
- Test how different models handle the same query
- Validate code snippets or reasoning outputs
FAQ from Multi LLM Cross-Check MCP Server
What are the prerequisites?
Python 3.8 or higher, API keys for the desired LLM providers, and the uv package manager installed via pip install uv.
How do I configure the server?
Set environment variables for each provider’s API key (OPENAI_API_KEY, ANTHROPIC_API_KEY, PERPLEXITY_API_KEY, GEMINI_API_KEY) in the claude_desktop_config.json file.
What if I don’t have a key for a certain provider?
That provider will be skipped automatically; you only need keys for the services you want to use.
How does error handling work?
Each LLM’s response is fetched independently. API errors are caught and returned in the response, and errors with one provider do not affect the others.
What transport does the server use?
—
