GuardCue

Connect it to the tool you built with

Once this is connected you never copy anything again. You say “check my app and fix what you find” inside the tool you already use, and it talks to us directly: what is wrong, the exact wording to fix it, then a second check to prove it worked.

The address you will paste

https://api.guardcue.com/api/v1/mcp

That is all you need. There is no key to generate, copy or keep safe, because you sign in with your email instead.

Before you start

  1. Connect your GitHub project to the site first. Without it we only see what a visitor sees, so problems inside your code are missed and the list is shorter. Connect it under Looking deeper. The answer always says when your code was not read.
  2. Open your tool in the site's project folder. That is the folder on your computer that holds this site's code, a copy of your GitHub project. Opened there, your tool can make each fix for you, with you watching. Opened anywhere else, it can still check your sites and say what is wrong, but it leaves your files alone. Built with Lovable? Skip this: Lovable already works on your project.
  3. Fixes go from your project folder to GitHub, then live. Your tool makes each fix in your project folder and sends it to GitHub, and your hosting puts it live from there. Before it changes anything, it switches to your main version and brings it up to date. It stops to ask if you have unsaved work, and never deletes or overwrites anything already on GitHub.
  4. A fix counts once it is live. Saving a fix in your project folder is not enough. Once it has reached GitHub and your site shows the change, ask whether it is fixed. That one problem is checked again straight away. Built with Lovable? Press Publish changes first.
  5. Your tool may stop and ask you first. Your tool has its own safety check, separate from us. In auto mode it can refuse steps that are hard to undo, such as deleting files or changing your database. That is your tool being careful, not a fault. It tells you what it wanted to do and why. To let it go ahead in Claude Code, press Shift+Tab to leave auto mode, say go, and approve each step. Or ask for the exact steps and do them yourself. Only approve a step you understand.

Words in a box are what you will see on their screen. You do not need to know what they mean, only to find them.

Claude Code

You build in a terminal with Claude Code.

  1. Run this one line:
  2. claude mcp add --transport http guardcue https://api.guardcue.com/api/v1/mcp
  3. Start Claude Code, type /mcp, choose guardcue, then Authenticate.
  4. Sign in with the email you use for GuardCue, then press Allow.

Lovable

You built your app by describing it to Lovable.

  1. In Lovable, open Connectors, choose Add custom, then MCP server.
  2. Name it GuardCue, and paste this address: https://api.guardcue.com/api/v1/mcp
  3. Leave the sign-in option on OAuth, and press Add & authorize.
  4. Sign in with the email you use for GuardCue, then press Allow.

Then, in your project's Lovable chat, type: Use GuardCue to check my site. To fix one, type: Fix the worst problem GuardCue found. A fix is live only after you open Publish and press Publish changes, so do that before GuardCue checks it.

Cursor

You write your app in Cursor.

  1. In Cursor, open Cursor Settings (the gear at the top right). In its left column, choose the icon with four squares. It opens a page called Customize.
  2. At the top, choose MCPs. Scroll to the bottom, press New MCP Server, and choose User, so it works in every project.
  3. Cursor opens a settings file called mcp.json. Replace everything in it with the text below, then save it (Cmd and S).
  4. {"mcpServers":{"guardcue":{"url":"https://api.guardcue.com/api/v1/mcp"}}}
  5. Go back to Customize. Find guardcue in the list and press Authenticate. Your browser opens.
  6. Sign in with the email you use for GuardCue, then press Allow. When your browser then asks to open Cursor, press Open Cursor.

Back in Cursor, guardcue shows as connected. Cursor files GuardCue under Reads, because it only reads. If Cursor asks before using it, you can allow it.

VS Code

You write your app in Visual Studio Code with Copilot.

  1. In VS Code, press Cmd, Shift and P together (Ctrl, Shift and P on Windows), type MCP: Add Server and choose it.
  2. Choose HTTP (HTTP or Server-Sent Events).
  3. Paste this address: https://api.guardcue.com/api/v1/mcp
  4. Where it asks for a Server ID, replace the suggested name with guardcue.
  5. Choose Global so it works in every folder. Then choose User Configuration, even though it says it is old, and not Copilot Global: User Configuration is the one VS Code chat reads.
  6. Start a new chat in Copilot, set it to Agent, and send a first message, such as: check my app with GuardCue.
  7. VS Code then asks whether GuardCue may sign you in. Press Allow and your browser opens. If it does not ask, open the list of servers (Cmd, Shift and P, then MCP: List Servers), choose guardcue and press Authenticate. Sign in with the email you use for GuardCue, then press Allow.

It starts when you send that first message, not before. Starting it from the list of servers before then shows an error, and that error means nothing. VS Code may also show a note asking you to **Review Migrations**: you can close it. If GuardCue ever stops answering, close VS Code and open it again.

Bolt

You built your app in bolt.new.

Coming soon. Until then, every problem on your Problems page comes with the exact words to paste into Bolt's chat, which does the same job by hand.

Replit

Your app was built and runs on Replit.

Coming soon. Until then, every problem on your Problems page comes with the exact words to paste into Replit's chat, which does the same job by hand.

Base44

You built your app in Base44.

Coming soon. Until then, every problem on your Problems page comes with the exact words to paste into Base44's chat, which does the same job by hand.

What you say once it is connected

Nothing to learn. You talk to the tool you already use, the way you already do:

What it can and cannot do

It reads. We tell your tool what is wrong and hand over the wording, and your tool makes the change with you watching. We never touch your app ourselves.

It also tells the truth about what it could not check. An agent handed a clean result will tell you that you are safe, so when something could not be looked at, the answer says so rather than reading as an all clear.

You stay in charge of what is connected. Every tool you allow is listed in your account, with a Disconnect button beside it.

If it does not connect

Tell us which tool you are using and what you saw, and we will walk you through it. One working day, usually much less. We would rather fix the step that confused you than answer the same email twice, so telling us is genuinely useful.

Ask us about connecting it

Connecting the tool you built with is part of the Solo and Business plans. It is not part of the Launch Check. See what each plan includes.

Connect GuardCue to the tool you built with · GuardCue