Claude Code

The agent this app was built against, and the only one it has been watched editing a live page.

Connect it

In the app, register with Claude Code. It writes a single Multiagentry entry into ~/.claude.json, under mcpServers.

Restart Claude Code and the tools appear. If it was already running, it will not see them until it restarts. That is Claude Code's behaviour, not ours.

What it writes, and what it does not

That file is yours: it holds your other MCP servers, your projects and your own settings. The app replaces only its own entry and treats every other section as opaque text it must not parse.

That is not a promise, it is how the writer is built, and it is checked against a copy of a real config file on every release, byte for byte, including what happens when you remove it. A backup of the file is written beside it before any change.

Maturity

Claude Code is the path this was developed against. It is the one that has been watched editing live pages through this app in every acceptance run since 1.30.0. Cursor and Codex register the same way and serve the same tools, but they have not had the same hours on them, and the app says so per agent rather than leaving you to find out.

How a conversation actually goes

Ask for the change in your own words. The agent picks a tool, the app makes the request, and what comes back is a draft on the real site. You are not handed a patch to apply. The work happened.

Before it plans anything on a site it has touched before, the agent can read back what has already been done there: pages created, sections finished, deployments, rollbacks. That is the project brain and the ledger, and it is why a new conversation does not start from nothing.