MCP 服务器

Arkheia Hallucination Detection MCP

Detect fabrication and hallucination in any LLM output. Score responses from GPT-4o, Claude, Gemini, Llama and 30+ models. Free tier included.

README

Arkheia MCP Server — Fabrication Detection for LLMs

Detect fabrication (hallucination) in any LLM output. Free tier included (1,500 detections/month).

Quick Start

1. Clone and install

git clone https://github.com/arkheiaai/arkheia-mcp.git ~/.arkheia-mcp
cd ~/.arkheia-mcp
python -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -r requirements.txt

2. Get an API key (free)

curl -X POST https://arkheia-proxy-production.up.railway.app/v1/provision \
  -H "Content-Type: application/json" \
  -d '{"email": "you@example.com"}'

3. Add to Claude Desktop

Edit your config file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "arkheia": {
      "command": "~/.arkheia-mcp/.venv/bin/python",
      "args": ["-m", "mcp_server.server"],
      "cwd": "~/.arkheia-mcp",
      "env": {
        "PYTHONPATH": "~/.arkheia-mcp",
        "ARKHEIA_API_KEY": "ak_live_YOUR_KEY_HERE"
      }
    }
  }
}

On Windows, replace ~/.arkheia-mcp with the full path using forward slashes (e.g. C:/Users/YourName/.arkheia-mcp) and use .venv/Scripts/python instead of .venv/bin/python.

4. Restart Claude and verify

Restart Claude Desktop (quit and reopen, not just close window). The Arkheia tools will appear automatically.

To verify, ask Claude: "Use arkheia_verify to check this response for fabrication: 'The Eiffel Tower is located in Berlin, Germany.'"

You should see a detection result with a risk level and confidence score.

Tools

Tool	Description
`arkheia_verify`	Score any model output for fabrication risk (LOW/MEDIUM/HIGH)
`arkheia_audit_log`	Review detection history
`run_grok`	Call xAI Grok + screen for fabrication
`run_gemini`	Call Google Gemini + screen for fabrication
`run_together`	Call Together AI (Kimi, DeepSeek) + screen
`run_ollama`	Call local Ollama model + screen
`memory_store` / `memory_retrieve` / `memory_relate`	Persistent knowledge graph