# ATP

- **Event:** [Built with Opus 4.6: a Claude Code hackathon](https://cerebralvalley.ai/e/claude-code-hackathon)
- **When:** Feb 10 at 12:00 PM – Feb 17 at 10:00 AM (EST)
- **Where:** Location TBA
- **Team:** [David Fant](https://cerebralvalley.ai/u/fant)
- **GitHub:** https://github.com/davidfant/agent-trigger-protocol
- **Demo video:** https://youtu.be/efYiuXhCcPc
- **Gallery:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery/137

MCP gave agents tools. ATP gives agents triggers.

Today, AI agents are purely reactive — they can't monitor your inbox, watch a Slack thread, or follow up on a PR review. ATP changes that.

ATP is an open, local-first protocol that lets agents subscribe to real-world events and act when they happen. Some examples:
Claude sends an email, a reply arrives three hours later, and Claude picks it back up.
A new Intercom message comes in. Claude investigates and replies. When the customer replies a day later, Claude picks it up in the same session.

It defines three roles: Watchers (adapters for Gmail, Slack, etc.), Core (a local daemon that routes events), and Hosts (AI agents like Claude Code).

Just as MCP enabled a community ecosystem of tool servers, ATP enables one for event watchers — making it easy to build async, long-running agents.

Our demo shows the full loop: Claude Code sends an email, subscribes to the thread, and resumes when a reply arrives — no human in the loop.

---

Markdown version of https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery/137. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt