MCP Setup Guide

Connect Claude Desktop to Lucairn in 2 minutes

1

Get your API keys

You need two keys: your DSA API key (from Lucairn) and your Anthropic API key (from console.anthropic.com). If you don't have a DSA key yet, sign up at /account/signup.

2

Open Claude Desktop settings

In Claude Desktop, go to Settings → Developer → Edit Config. This opens your claude_desktop_config.json file.

3

Add the MCP server

Paste the configuration below into your claude_desktop_config.json file. Replace the placeholder values with your actual keys.

4

Restart Claude Desktop

Close and reopen Claude Desktop. You should see "lucairn" in the MCP servers list. Every message you send now routes through Lucairn's privacy pipeline.

claude_desktop_config.json
{
  "mcpServers": {
    "lucairn": {
      "command": "npx",
      "args": ["-y", "@lucairn/mcp-server"],
      "env": {
        "DSA_GATEWAY_URL": "https://gateway.lucairn.eu",
        "DSA_API_KEY": "<your DSA API key>",
        "ANTHROPIC_API_KEY": "<your Anthropic API key>"
      }
    }
  }
}

How it works

  • Every message is scanned for PII (names, emails, addresses, medical terms) before it reaches the LLM.

  • Detected PII is replaced with safe placeholders. The LLM never sees real personal data.

  • Developer-tier responses contain placeholders so your code never receives raw PII. Pro and Enterprise tiers can enable automatic re-linking back to the original values.

  • A Lucairn Certificate is generated for each request — cryptographic proof of what was sanitized.

Capability matrix

Claude Desktop uses MCP over stdio (request/response), so streaming is not a separate concern at the client boundary. The Lucairn pipeline still has real gaps around multimodal and tool-call payloads — read the table before relying on it for sensitive workflows.

  • Text messages via dsa_messages / dsa_analyze

    Full PII sanitisation on every tool call, signed Lucairn Certificate per request.

  • System prompts

    Sanitised end-to-end alongside the user content.

  • Multi-turn conversations

    Each tool call is a fresh request through the pipeline; one certificate per call.

  • Streaming responses

    !

    Not applicable: MCP stdio is request/response. Claude Desktop renders the full reply once the tool call returns. No SSE involved.

  • Anthropic tool-calls / function calling inside the prompt

    ✕ Roadmap

    Tool definitions and tool-call arguments forwarded to Anthropic are not sanitised today. Keep PII out of any tools you pass through dsa_messages until per-call DLP for tools ships.

  • Image / file / audio attachments

    ✕ Roadmap

    The MCP server only forwards text content. Multimodal inputs have no Lucairn pipeline coverage.

  • Anthropic prompt caching

    ✕ Roadmap

    Each request is processed independently so the per-call evidence stays valid. No cache reuse across requests.

Tool-call DLP and multimodal sanitisation are tracked on the roadmap. Subscribe to the changelog for ship dates. Read the changelog.

Want to see this in action?

Book a working session — we'll walk through your use case together.