MCP server for the EU AI Act: integrate AI Act Radar in Cursor and Claude Desktop
As of 18 November 2026 · checked against the MCP server code (version 0.2.0)
Information only, not legal advice. The MCP tools surface published events and references. Confirm any concrete decision with qualified counsel.
AI Act Radar runs a hosted Model Context Protocol server at https://mcp.aiactradar.com/mcp/v1. It exposes the event feed, the obligation matrix and article lookups as tools any MCP client can call. Instead of pasting regulation text into a chat, your editor or assistant queries the server directly. This page lists the tools, shows the configuration, and gives three prompt patterns.
Tools exposed
| Tool | Purpose | Returns |
|---|---|---|
query_ai_act_updates | Filtered events | Events by since, source, topics and minimum severity |
get_obligation_for_role | What applies to a role and risk class | Articles, obligations, dates and sources from the obligation matrix |
summarize_changes_since | Digest since a date | Markdown digest grouped by severity |
lookup_article | Events for one provision, like "Art. 53" | Events that reference that Article, Annex or Recital |
Cursor setup
Open Cursor Settings, navigate to MCP, and add a remote server:
{
"mcpServers": {
"ai-act-radar": {
"url": "https://mcp.aiactradar.com/mcp/v1"
}
}
} Without a key the server answers in a read-only trial mode: at most 10 items per result and 40 calls per hour per IP. A free beta key (sent as an Authorization: Bearer header) removes both limits; a wrong key is rejected with 401. Details in the Cursor guide.
Claude, Claude Code, VS Code, ChatGPT
Any client that accepts a remote MCP server URL can use the server. Claude Desktop and claude.ai add it as a custom connector (guide). Claude Code uses claude mcp add --transport http (guide). There are also guides for VS Code and ChatGPT. Clients that only speak stdio can use the mcp-remote bridge. We do not publish an npm package for a local server.
Three concrete prompt patterns
The point of MCP is not to replace the API but to make ad-hoc reasoning fast. Three patterns we use ourselves:
- "What changed since I last looked?" Prompt: "Summarise high-severity events on transparency since 2026-11-01." The agent calls
summarize_changes_sincewith the date and topic. - "What do I owe?" Prompt: "I deploy a high-risk system. What applies and from when?" The agent calls
get_obligation_for_rolewith role and risk class. - "Walk me through this article." Prompt: "What has happened around Art. 50 recently?" The agent calls
lookup_article.
Operational notes
- The MCP server is read-only. Webhook subscription management remains on the REST API.
- Results are capped with a
limitargument where the tool has one. Without a key the cap is 10 items. - Discovery document:
GET https://mcp.aiactradar.com/.well-known/mcp.json. Protocol: JSON-RPC 2.0 over HTTP POST.
Tools reference and Cursor recipes
Full schemas live at docs.aiactradar.com/mcp/tools. Cursor setup and a first prompt are documented at docs.aiactradar.com/mcp/cursor.
→ See live updates: https://aiactradar.com
→ Subscribe to webhook: https://docs.aiactradar.com/quickstart