Reachable by your AI assistant

More and more of the setup work in a small app doesn't happen in a dashboard anymore. It happens in a conversation — with Claude, with Cursor, with whatever assistant is already open in the editor. You describe what you want, and it wires it up.

If that's where the work happens, a support tool should be reachable from there. So we made presence reachable: your assistant can set up your contact form without you copying keys between tabs.

One conversation, from nothing to a live form

There's a remote MCP server at https://usepresence.com/mcp. Connecting it takes one line:

claude mcp add --transport http presence https://usepresence.com/mcp

Your browser opens, you sign in — or create an account right there — and you're connected. No API key to find, copy, or paste; OAuth handles it in the browser. New accounts are provisioned with a team and a publishable key automatically, so an agent can go from "no account" to "contact form shipped" inside a single conversation.

From there your assistant has seven tools: it can list and create teams, list and create sources, pull your setup instructions, and read your inbox — list conversations and open any one of them.

Where we drew the line

Your agent can set everything up and read what's coming in. It cannot reply to a customer, and it cannot revoke a key. Those stay in the web app, done by a person.

It's fine for a machine to read a customer's message and get the plumbing ready. Answering that customer is still yours to do. The same line runs through our AI-drafted replies: the model writes a first pass, you send it.

Docs your assistant can actually read

We also made the words machine-readable. There's an /llms.txt and a fuller /llms-full.txt, every doc and post is available as raw markdown (just add .md), and pages carry structured JSON-LD. So when you ask an assistant how presence works, it can read our docs directly and cite them, instead of guessing from stale training data.

The full walkthrough is in Connecting via MCP.