Publish your app as an MCP server

1. 1

   Open your app in Lovable and go to 'More' then 'Agent integrations'.

   This is where you manage the connection between your app and AI assistants.

2. 2

   Select 'Enable agent integrations'.

   This tells Lovable to create the special server and suggest actions your app can perform for AI assistants.

3. 3

   Choose who can use your app's actions: either 'Protected with OAuth' (users sign in) or 'Public — no login' (anyone can use them).

   This sets the security level for your app's actions when used by AI assistants. 'Protected' is usually safer.

4. 4

   Review the suggested actions (called 'tools') in the 'AI tools' section and ask Lovable to change them if needed (e.g., add, remove, rename).

   You need to make sure the AI assistant can only do what you intend and that the descriptions are clear.

5. 5

   Publish your app.

   This makes your app's actions live and available for AI assistants to use, and generates the link you'll share.

6. 6

   In 'More' then 'Agent integrations', copy the link from the 'Your MCP link' card.

   This is the link your users will add to their AI assistants to connect your app.

7. 7

   Test your app's connection and actions in an AI assistant like ChatGPT or Claude.

   It's important to confirm everything works as expected before sharing the link with others.

Add Agent integrations to this app so users can connect it to AI assistants like ChatGPT and Claude. I want to make sure users have to sign in before they can use the tools.

MCP server

A special server that acts as a translator, allowing AI assistants to understand and use the actions your app can perform.

Tools

These are the specific actions or features from your app that an AI assistant can call or use, like 'look up an order' or 'create a request'.

Agent integrations

The way your Lovable app connects to and works with AI assistants, letting them use your app's features.

OAuth

A secure way for users to give an AI assistant permission to use your app on their behalf, without sharing their password directly with the assistant.

• Exposing sensitive data or actions that are slow, expensive, or destructive without proper sign-in or usage limits.
• Not reviewing each 'tool' carefully to ensure it only does what's intended and returns only necessary data.
• Assuming interface-level controls (like hidden buttons) protect backend actions from AI assistant calls.
• Publishing an app to a workspace with public (no sign-in) tools, which will be blocked.
• Forgetting that tool calls consume resources and incur charges just like direct app usage, without built-in rate limits.
• Not testing the full connection flow and individual tools before sharing the MCP link with users.