MCP setup

Create an API key (ttk_live_…).
Add TuringTap to your MCP client — pick yours:

Claude Code (CLI)

One command:

claude mcp add --transport sse turingtap https://mcp.turingtap.ai/sse \
  --header "Authorization: Bearer ttk_live_xxx"

Verify with /mcp inside a session — you should see turingtap connected and its tools listed.

Claude Desktop

Claude Desktop launches local processes, so bridge to our SSE endpoint with mcp-remote. Settings → Developer → Edit Config (opens claude_desktop_config.json; on macOS it lives at ~/Library/Application Support/Claude/) and add:

{
  "mcpServers": {
    "turingtap": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote",
        "https://mcp.turingtap.ai/sse",
        "--header", "Authorization:Bearer ttk_live_xxx"
      ]
    }
  }
}

Save, then fully quit and reopen Claude Desktop. The tools appear under the 🔨 icon.

Cursor / Windsurf / other SSE-native clients

{
  "mcpServers": {
    "turingtap": {
      "type": "sse",
      "url": "https://mcp.turingtap.ai/sse",
      "headers": { "Authorization": "Bearer ttk_live_xxx" }
    }
  }
}

(Cursor: .cursor/mcp.json in your project or ~/.cursor/mcp.json globally.)

Any of these is enough for observe-only — no other install required. Give your agent the SKILL.md so it knows the workflow patterns.

No-install path (observe & analyze)

The desktop agent is required for goto()/act(), screencast handoffs, and LAN targets — but not for reading the wire. session_info() returns the proxy host, port, a per-session token, and the CA certificate; your AI can configure any HTTP client with those and every request it (or you) sends through the proxy shows up in get_traffic() / analyze():

# example: route curl through your session
curl -x http://<session-token>:x@proxy.turingtap.ai:8443 \
     --cacert turingtap-ca.pem \
     https://api.example.com/v1/things

Works the same for httpx, Playwright you already run, or a whole test suite — point it at the proxy and let your AI watch.

Full capability (with the agent)

Install turingtap-agent and paste your key into ~/.turingtap/agent.toml. This adds:
- goto() / act() — your AI drives a real Chromium on your machine
- ask_human() screencasts — solvable in any browser or the mobile app
- LAN targets (Pro, off by default — enable in Settings)

The agent also exposes a local SSE endpoint that forwards to the cloud and adds agent_status / agent_lan_test:

{
  "mcpServers": {
    "turingtap": {
      "type": "sse",
      "url": "http://localhost:7847/sse"
    }
  }
}

Once connected, call session_info() to get everything your AI needs for the session.