MCP Server

What you get

The Lighthouse MCP server exposes three tools to your AI client. The model calls them on your behalf when the conversation is about your waitlist.

• list_waitlists — every waitlist on the account with id, name, slug, goal, and current signup count.
• get_waitlist — full detail for one waitlist including any survey questions attached to the signup form.
• list_signups — signups for a waitlist (email, note, survey answers, created_at) with pagination.

More tools land here as they ship. Nothing to update on your end: the client refetches the tool list on every session.

1. Get your Pro API key

The MCP server reuses the same Bearer API keys the REST API uses. Generate one from the dashboard:

• Sign in and open Settings → API keys
• Click Create key, name it something you will recognise (e.g. “Claude Desktop”)
• Copy the lh_... value that appears. It shows only once

Keys are Pro-only and rate-limited to 60 requests/minute per key.

2. Wire it into your client

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or the equivalent path on your OS. Add the entry below (merge with anything already there). Restart Claude Desktop.

{
  "mcpServers": {
    "lighthouse": {
      "url": "https://lighthouse.build/api/mcp",
      "headers": {
        "Authorization": "Bearer lh_..."
      }
    }
  }
}

Cursor

Open Cursor Settings → MCP and add an entry that points at the same URL and header. The JSON body is identical to Claude Desktop's:

{
  "mcpServers": {
    "lighthouse": {
      "url": "https://lighthouse.build/api/mcp",
      "headers": {
        "Authorization": "Bearer lh_..."
      }
    }
  }
}

ChatGPT + other MCP clients

Any client that speaks streamable-HTTP MCP works. Give it the URL https://lighthouse.build/api/mcp and an Authorization header with your Pro API key. That is the whole config.

3. Try it

Prompts that just work once the server is connected:

• “What waitlists do I have on Lighthouse?”
• “Show me the last 20 signups on the Runwell waitlist. What are the common themes in their survey answers?”
• “Group my Runwell signups by what tool they said they use today and count each group.”
• “Draft a launch email to the folks who said price was their biggest concern.”

Testing the endpoint by hand

The MCP server is a plain JSON-RPC 2.0 POST endpoint. You can hit it with curl to verify auth and the tool list:

Initialize

curl -X POST https://lighthouse.build/api/mcp \
  -H "Authorization: Bearer lh_..." \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{}}'

List tools

curl -X POST https://lighthouse.build/api/mcp \
  -H "Authorization: Bearer lh_..." \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/list"}'

Call list_waitlists

curl -X POST https://lighthouse.build/api/mcp \
  -H "Authorization: Bearer lh_..." \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc":"2.0",
    "id":3,
    "method":"tools/call",
    "params":{"name":"list_waitlists","arguments":{}}
  }'

Limits + FAQ

Is my data safe?

The API key gates every call. Rotate or delete keys from the dashboard at any time. The MCP server only exposes data on your account: no cross-tenant reads are possible because ownership is verified on every tool call.

What does the AI see?

Exactly what the tool call returns and nothing else. There is no background upload or ambient sync; the AI queries only what the user asks it to.

Rate limits

60 requests per minute per API key, shared across REST and MCP. Every response carries X-RateLimit-Remaining so clients can pace themselves.

What happens if I downgrade from Pro?

The API key keeps existing but every MCP call returns 403 until you are back on Pro. Your data is untouched.