legalithm-mcp-server

EU AI Act compliance tools for Claude Code and Cursor, over MCP (stdio). classify, explain_obligation, and generate_disclosure run fully offline (the rule engine is bundled — no network, no API key); check_record reads the public Trust Center.

Install

One-liner (wires both editors + the CLI): npx legalithm setup

Cursor — click the Add to Cursor badge above, or add to .cursor/mcp.json.

Claude Code (.mcp.json in your repo, or claude_desktop_config.json):

{
  "mcpServers": {
    "legalithm": { "command": "npx", "args": ["-y", "legalithm-mcp-server"] }
  }
}

Cursor (.cursor/mcp.json): same block as above.

Then drop the editor rules into your repo so the agent knows to use the tools:

• Cursor: copy templates/legalithm-eu-ai-act.mdc → .cursor/rules/
• Claude Code: append templates/CLAUDE.md to your CLAUDE.md

Tools

| Tool | Mode | Purpose | |---|---|---| | classify | offline | Risk tier (unacceptable/high/limited/minimal) + cited rationale for an AI use case | | explain_obligation | offline | The obligations for a role + risk tier, each with its Article citation | | generate_disclosure | offline | Article 50 transparency snippet (chatbot / genai-content / deepfake / emotion), EN or DE | | check_record | online | Fetch a published Trust Center compliance record by org slug |

Set LEGALITHM_API_URL to point check_record at a non-default host.

Output is checked against Regulation (EU) 2024/1689 — not legal advice.