Connect AIR to Claude and other MCP clients
Run AIR risk assessments from Claude, Cursor, VS Code and other AI assistants through MCP.
· 1 min read
AIR has an MCP server. MCP, the Model Context Protocol, lets an AI assistant use outside tools. With the AIR server, your assistant can find projects, upload evidence, run assessments and read the reports.
Two ways to connect
- Local, for coding assistants such as Claude Code, Cursor, VS Code and Windsurf. The server runs on your computer from the npm package @thalus-ai/mcp-air. The assessments still run in the AIR cloud.
- Hosted, for Claude and other remote MCP clients, at https://mcp.air.thalus.ai/mcp. You sign in through the AIR portal.
Local setup
- In the AIR portal, open your domain, go to API Keys, and create a key.
- Add the server to your client. For Claude Code, run:
claude mcp add --env AIR_API_KEY=… --transport stdio air -- npx -y @thalus-ai/mcp-air@1.5.0
Other clients use the same npx command in their MCP settings file. You need Node.js 20 or later.
- Ask your assistant to list your AIR domains and projects to check the connection.
No account yet? The server starts without a key, and you can create an account inside the conversation. You type your details into a form; the model does not see them.
API key access levels
- Assessment runner: list domains and projects, start assessments, and read reports on documents already uploaded.
- Full pipeline: everything above, plus upload documents, create projects, search, and see the domain portfolio.
What your assistant can do
- Find domains and projects, and create a project.
- Upload documents and wait until AIR has read them.
- Start an assessment, follow its steps, and read the summary or the full report.
- Answer the open EU AI Act questions in a short form.
- Record a decision, such as "this topic does not apply", with a reason.
- Check credits and ask for more.
The full setup guide is in the AIR documentation.