Most product teams already plan launches with an AI assistant. They draft release notes in Claude, think through onboarding in ChatGPT and write code in Cursor. Then someone still has to open another tool and build the tour by hand.

Cueston now runs a Model Context Protocol (MCP) server, so that last step happens where you already work. You describe what you want in plain words. The assistant looks at your real pages, drafts the guidance in your Cueston workspace, checks every element it points at and sends you a preview link. Nothing goes live until you say so.

What you can ask for

• "Announce the new scheduled reports feature: a What's new post and a tooltip on the Schedule button."
• "Build a five-step tour of the invoice flow for new admins and send me a preview link."
• "Which tours broke after last week's release? Repoint them and show me before you publish."
• "How did the onboarding checklist do in the last 30 days, and where do people drop off?"

From a sentence to a preview link

The assistant works in the same order a careful product manager would. First it learns your product: your workspace rules, the pages Cueston knows, UI elements with selectors it has checked, your events and saved audiences. Then it drafts the tour, tooltip, banner, checklist, survey or announcement. Before it shows you anything, it validates the draft against the real page and the size of the audience. Last, it creates a preview link that shows the draft on your own site.

When you are happy, it publishes or asks a teammate for review, depending on your approval rules. Afterwards it can read analytics, funnels and survey results, and find guidance that broke after a release.

Connect it in a minute

Add Cueston as a connector with this URL. Your app opens Cueston so you can sign in, pick the workspace and choose what the assistant may do. There are no keys to copy.

https://app.cueston.com/api/mcp

• Claude (web or desktop): Settings → Connectors → Add custom connector.
• ChatGPT: Settings → Connectors, with Developer mode on if it is offered.
• Cursor and VS Code: add the URL to mcp.json and sign in on first use.
• Any other MCP app: give it the URL and it finds the sign-in page on its own.

In Claude Code it is one command:

claude mcp add --transport http cueston https://app.cueston.com/api/mcp

Or paste one prompt and let the agent set itself up

Coding agents such as Claude Code, Cursor, Codex or Windsurf can add Cueston to their own settings. Paste this into the chat:

Add the Cueston MCP server to your MCP settings.

Name: cueston
URL: https://app.cueston.com/api/mcp (Streamable HTTP)
Sign-in: OAuth. There's no API key to copy; I'll sign in to Cueston in my browser.

If you can only run local (stdio) servers, use the npm package instead:
command "npx", args ["-y", "cueston", "mcp"]. I'll give you an API key for CUESTON_API_KEY.

When it's added, tell me how to finish signing in. Then call get_workspace and tell me what you see.

On npm for apps that only run local servers

Some MCP apps can only start local (stdio) servers. For those, the cueston package on npm is the bridge. It forwards every request to the hosted server with your API key, so you get the same tools and the same rules:

{
  "mcpServers": {
    "cueston": {
      "command": "npx",
      "args": ["-y", "cueston", "mcp"],
      "env": { "CUESTON_API_KEY": "sk_…" }
    }
  }
}

The same package is the Cueston CLI, for building and publishing guidance from your terminal or CI.

Your first request

Once you are connected, paste this to see the whole loop on your own product:

Use the Cueston tools.

1. Look at my workspace, the pages Cueston knows and the guidance that's already live.
2. Suggest three pieces of guidance that would help new users most, and say why.
3. Build the first one as a draft, with selectors you've checked on the real page.
4. Validate it and send me a preview link. Don't publish until I say so.

You stay in control

• When you connect, you choose one level: read only, create and edit drafts, or edit and publish.
• Everything the assistant builds is a draft until someone publishes it, and your approval rules still apply.
• It can never do more than you can, and it never touches your team, billing, security settings or keys.
• Every action is in the audit log, labelled with the name of the app that did it.
• You can see and disconnect connected apps at any time in Settings → Integrations → Connected apps.

Setup steps for each app, the full tool catalog and ready-made prompts are in the MCP documentation. If you use Cueston from the dashboard and want the short version, read Work with Claude or ChatGPT in the help center.