# TuringTap

> TuringTap is an MCP server that gives AI agents a live, decrypted view of HTTP(S) traffic plus a human-in-the-loop handoff for the parts an agent cannot do alone (2FA, captchas, logins, consent screens). The user's browser or HTTP client is proxied through a cloud mitmproxy; the agent reads the wire with get_traffic()/analyze(), drives a real Chromium with goto()/act(), and hands off to the human with ask_human(). Session traffic lives in per-session RAM only and is purged on close -- zero traffic-data retention, no telemetry. Source-available under BSL 1.1.

Remote MCP endpoint: https://mcp.turingtap.ai/sse (SSE, Authorization: Bearer ttk_live_... API key).

## Docs

- [SKILL.md](https://turingtap.ai/SKILL.md): canonical agent skill doc -- all 8 tools, worked patterns, credential hygiene rules
- [Agent quickstart](https://turingtap.ai/docs/quickstart): numbered, agent-executable setup: key -> MCP config -> session_info() -> proxied curl -> get_traffic() -> analyze()
- [MCP setup](https://turingtap.ai/docs/mcp-setup): per-client setup (Claude Code, Claude Desktop, Cursor/Windsurf) including the no-install observe-only path
- [Docs overview](https://turingtap.ai/docs): what TuringTap is and is not, the 8 MCP tools

## Legal

- [Terms of service](https://turingtap.ai/legal/tos)
- [Privacy policy](https://turingtap.ai/legal/privacy)

## Source

- [GitHub](https://github.com/TuringTap): source-available (BSL 1.1, audit-only)