Let ChatGPT and Claude use your app, safely
Your team already asks ChatGPT and Claude to do work. With MCP, the Model Context Protocol, those assistants can use your app directly: look up a booking, draft replies or update a record. The hard part is doing that without giving an AI the keys to everything.
An agent acts as a person, never above one
Every Tervane app serves its own MCP endpoint from your Vercel deployment. A member connects an assistant, by OAuth from ChatGPT or claude.ai or with a key they create, and the agent can only do what that member could do. The same database rules apply.
Risky actions wait for a person
Actions you mark as needing approval do not run when an agent asks. They queue in the app's approval inbox, and a member approves or rejects them.
Everything is logged
The activity log records who did what, and whether it was a person or which agent. Keys can be revoked at any time.
Using your own AI key
AI features inside the app, like drafting replies or sorting requests, run on your own OpenAI key and land in the same approval queue.