Adds fabrication detection directly into your Claude workflow with the arkheia_verify tool that scores any LLM output for hallucination risk. Beyond verification, you get wrapped inference tools for Grok, Gemini, Together AI, and Ollama that automatically screen responses before returning them. Useful when you're chaining multiple model calls or need audit trails of what passed and failed detection. Includes a basic memory graph for persisting knowledge across sessions. Free tier covers 1,500 detections monthly. Requires a quick API key provision via curl and drops into Claude Desktop config with Python 3.10+.
claude mcp add --transport stdio io.github.arkheiaai-mcp-server -- npx -y @arkheia/mcp-server