DEV Community

JustOneCX
JustOneCX

Posted on Originally published at justonecx.com

Use an AI assistant for customer onboarding with MCP

You can use an AI coding assistant for customer onboarding by connecting it to the JustOneCX remote MCP server at https://api.justonecx.com/mcp. It works with Claude Code, Codex, Cursor and Gemini CLI. You sign in through your browser, pick the workspace and choose an access level. Everything it creates is a draft until a person publishes it.

What does the assistant do for onboarding?

Once connected, the assistant can install the SDK, build product tours from your code, announce features, add surveys, answer chat, manage your roadmap, read your analytics and event log, set up funnels and Brand Kit, and find out why something does not show. It acts as you, so it can never do more than your own role allows.

For a vague request such as "add a tour", a planning skill makes it study your product and propose a ranked list first. It chooses where each tooltip, survey or announcement sits from where the element is on screen, and asks when unsure.

How do you connect it?

Your assistant signs in through your browser the first time. No token goes in any config file.

Claude Code

claude mcp add --transport http justonecx https://api.justonecx.com/mcp
Enter fullscreen mode Exit fullscreen mode

Then run /mcp, pick justonecx and sign in.

Codex. Add this to ~/.codex/config.toml, then run codex mcp login justonecx:

[mcp_servers.justonecx]
url = "https://api.justonecx.com/mcp"
Enter fullscreen mode Exit fullscreen mode

Cursor. Put this in .cursor/mcp.json:

{ "mcpServers": { "justonecx": { "url": "https://api.justonecx.com/mcp" } } }
Enter fullscreen mode Exit fullscreen mode

Gemini CLI. Add this to ~/.gemini/settings.json, then run /mcp auth justonecx:

{ "mcpServers": { "justonecx": { "httpUrl": "https://api.justonecx.com/mcp" } } }
Enter fullscreen mode Exit fullscreen mode

For CI and tools without browser sign-in, you can create a token in Settings → Developers → AI assistants. Keep the token in an environment variable and never in a config file. A token has the same limits as the browser sign-in and can be revoked in the same place. See the MCP server docs.

What are the access levels?

When you sign in, you pick how much the assistant may do.

Level What it can do
See your content and setup Reads only.
Create and edit drafts (default) Nothing your users see changes until a person publishes.
Make changes live Can also publish. Off unless you choose it.

A workspace owner can switch AI assistants off or cap them in Settings → Developers → AI assistants.

What can it not do?

It cannot empty the Trash, export survey responses as a file, or change members, roles, billing or keys. For those it points you to the dashboard page and the steps.

It never handles the Secret key. The Secret key is never returned, read, asked for or pasted into chat. You set JUSTONECX_SECRET_KEY in your own server environment yourself. The get_install_info tool returns your Public key and the install snippet, never the Secret key.

For identity verification, it adds the server code and an .env.example entry, and passes userHash to identify(). It switches Require on only after your live site sends userHash and you say yes, because that setting changes the live site. See identify users with the SDK.

What stays a draft, and what needs confirm?

Everything it creates is a draft until a person publishes it. Changes that reach customers or the live site need an explicit confirm. Examples are:

  • A chat reply, which goes to the customer at once.
  • A roadmap status change, because voters are emailed.
  • Workspace settings and custom CSS.
  • Deleting a tour, announcement or survey, which leaves the site at once.

Contact details such as email, phone and address are hidden unless the task asks for them. Every change shows in your audit log as made by the AI assistant.

How should you review the assistant's work?

Use the same steps every time.

  1. Ask it to run workspace_status. It returns your role, plan and a setup checklist, with a fix for every failing item.
  2. Ask it to build the content as a draft, for example a tour.
  3. Run validate_tour to check the selectors, page rule and step flow.
  4. Open the preview link. get_tour_preview_link previews the draft on your site for 24 hours.
  5. Read the draft in the dashboard. Check the copy and the targeting.
  6. Publish it yourself, or give a clear yes to publish through the assistant at the "Make changes live" level.

If the tour does not show after publishing, use the product tour not showing checklist. For a hands-on walk-through with one assistant, read add a product tour with Claude Code.

Can it help beyond tours?

Yes. It can list ideas on your roadmap, read the moderation queue and check that an event arrived with the events tool, which is read-only. Analytics tools return aggregated numbers and no person ids, emails or comment text. Surveys, announcements and What's new entries are saved as drafts the same way.

Which setup facts should you check yourself?

Three things always stay with you. Set the Secret key in your server environment. Review each draft before it goes live. Decide when to switch on Require for identity verification.

Frequently asked questions

Which assistants work with the MCP server?

Claude Code, Codex, Cursor and Gemini CLI. Chat-based builders that accept a remote MCP server work the same way, with the URL and then OAuth or a bearer token.

Can the assistant publish without asking me?

Only if you choose the "Make changes live" level. At the default level it creates and edits drafts only. Changes that reach customers also need an explicit confirm.

Does the assistant see my Secret key?

No. It never handles the Secret key. You paste it into your own server environment yourself.

Can I limit what assistants do in my workspace?

Yes. A workspace owner can switch AI assistants off or cap them in Settings → Developers → AI assistants.

See how it works in the live demo.

Top comments (0)