Connect with AI
Paste one prompt into the AI tool that has your site's code open; the agent connects the site and you approve in the browser.
The fastest way to connect a site: one prompt, copied from Mapled and pasted into the AI tool that has the site's code open. The agent does the work in that conversation; you approve three things in the browser — the connection, the content model, and anything destructive.
What you need
- A Mapled project (Create a project; Blank is the usual start for an existing site).
- The site's code in a git repository, opened in an AI tool that can read and edit files and run commands: Claude Code, Codex or Cursor.
- Node 18 or newer where the agent runs commands.
A chat without the site's files can still model content through Mapled's MCP server, but it can't wire the site — the agent says so right away.
The prompt
In Mapled, open the project and press Connect with AI — on the Overview, in Integrations, or Get the prompt at the end of New project — then Copy prompt. The prompt names your project by its id and gives the agent one command to start with:
npx -y @mapled/cli connect --project <id>
It also tells the agent how to work: do everything itself, never ask you for a token or a key, write what it puts into Mapled for people in English, and handle branches, pull requests and the merge into main on GitHub without asking again.
What happens next
- Connect. The command opens Mapled in your browser on that project. Press Authorize. The connection shows up in Integrations, where you can revoke it at any time. Nothing is copied or pasted.
- Plan. The agent reads the code, finds the hardcoded texts, images and lists, and proposes one plan: the collections and singles with their fields, the records to import, the files it will change. You get an email, and the plan opens in Mapled. Review it and press Approve plan — or Reject, and the agent proposes a better one.
- Apply and wire. The agent sees your decision by itself and carries on: Mapled creates the content model and imports the content; the agent installs the SDK, replaces the hardcoded content with reads, builds the site and checks it with
npx -y @mapled/cli doctor. - Ship. The agent commits, pushes and merges into the branch the site deploys from, and leaves
MAPLED.mdin the repository for whoever works on the site next. - Verify. Mapled checks the deployed site itself: that it reads content, that publishing reaches it, that preview works. The run's page in Integrations shows the checklist.
- Publish. The agent never publishes. Open Review & publish and press Publish now — your site now reads its content from Mapled.
How AI setup works explains each step, the checks and what the agent may and may not do.
Give your agent Mapled for good
The prompt is all a first connection needs. To make Mapled known to your agent in every project, install the mapled skill once — the Add Mapled to your agent card on the dashboard copies the command:
npx skills add mapledhq/agent-kit -g
It adds the skill to the AI agents it finds on your computer. Afterwards, asking an agent to "connect this site to Mapled" is enough.
- Connect with Claude Code · Paste the prompt from Mapled into Claude Code opened on your site's repository, then approve in the browser.
- Connect with Codex · Paste the prompt from Mapled into Codex opened on your site's repository, then approve in the browser.
- Connect with Cursor · Paste the prompt from Mapled into Cursor's agent with your site's repository open, then approve in the browser.
- How AI setup works · The connection, the plan you approve, the checks Mapled runs on your site, and what an agent may and may not do.