Your coding agent can set up your product’s AI agent now

Adding an AI agent to a product used to be a context switch. You read your own code to work out what the agent should be allowed to do, opened a dashboard in another tab, created something, copied a key, came back to your editor, pasted it, went back to check it worked. None of it was hard. All of it was the kind of work that gets put off until next week.

The coding agent you already have open knows your codebase better than any dashboard ever will. So now it does the setup too.

One sentence instead of a checklist

Connect Verb to your coding agent once, then say what you want in the same place you say everything else:

add Verb to this app

From there your agent works through it the way you would, only without the tab switching. It reads your app and writes down the things your users could ask for: look up an order, change a plan, invite a teammate, export last month. It shows you that list before anything else happens, because that list is what an AI will be allowed to do inside your product, and you should see it the way you see any other change: as a diff.

Then it creates your site, puts those actions on a test site that only runs on your own computer, adds the one line your app needs, and tells you what is left. You open your app on localhost and ask it something.

The two moments it stops and asks

Most of setup is safe to just do, and your agent does it. Two steps change what the outside world sees, and those stop:

• Going live on your domain. Before your real users can reach the agent, your coding agent tells you exactly which domain is about to go live and waits for a yes.
• Letting actions run for real on your test site. Until you turn this on, anything that would change data is simulated. Your agent tells you what turning it on means before it does it, and if none of your actions change anything, it says so and does not waste your time asking.

Everything your users later ask the AI agent to change still asks them for a yes, with the real details on screen. Setting it up from your editor changes who does the setup, not how carefully your product behaves.

What it does not do

It cannot read your users’ conversations. Those are your customers’ words, and they stay in the Verb dashboard where you chose to look at them. It cannot delete a site. And it cannot do anything the dashboard would not let you do yourself, because it is working as you.

Works with what you already use

Claude Code, Codex, Cursor, VS Code and Windsurf all connect with one command or one short config block. So does the Claude app, and ChatGPT with developer mode on, if you would rather talk it through than work in an editor. For Claude Code it is this:

claude mcp add --transport http --scope user verb https://app.askverb.com/api/mcp

The first time your agent uses it, a browser tab opens. You sign in with the same email code as the Verb dashboard and approve that editor by name. There is no API key to copy, store or rotate.

Try it on a real project

Building and testing is free and needs no card. Setup for every editor is at askverb.com/docs/mcp, and if you want to see the whole thing as a conversation first, the next post is the five prompts from an empty account to a working agent.