Lovable: WebMCP Lets Browser Agents Use Your App
Lovable now lets apps expose WebMCP tools so an in-browser AI agent can act on the page a person has open. With agent integrations enabled, the tools run in the visitor's own tab using their signed-in session. Lovable starts with read-only or easily undone actions and adds destructive ones only on request.
Key Takeaways
- WebMCP tools let an in-browser AI agent act directly on the page a person has open in a Lovable-built app.
- The tools run in the visitor's own browser tab with their signed-in session, not on a separate server.
- WebMCP tools are separate from the app's MCP server, so builders can use either or both.
- Lovable defaults to read-only or easily undone tools, limiting what an agent can do out of the box.
- Actions that delete, pay, send, publish, or share are added only when the builder asks for them.
- Builders can ask Lovable in project chat to add the tools without setting up a server.
What Changed
Lovable added support for letting browser agents use the apps people build on the platform. When agent integrations are enabled for an app, Lovable adds WebMCP tools to it. A browser-based AI agent can then act on the page a person currently has open, for example by marking a task as done or filtering a list, instead of clicking through the interface.
How the Tools Run
The WebMCP tools run in the person's own browser tab and use that person's signed-in session. They are separate from the app's MCP server, so an app can offer one, the other, or both. Builders who do not want to run a server can also ask Lovable in project chat to add the browser tools on their own.
Safety Defaults
Lovable begins with tools that only read information or that are easy to undo. Tools that delete data, take payments, send messages, publish content, or share anything are added only when the builder explicitly asks for them. This keeps a newly agent-ready app from handing risky actions to an agent by default.
Why It Matters
Agents that rely on screenshots or page scraping are fragile and break when a layout changes. By exposing named actions tied to app logic, Lovable apps become easier for browser agents to operate reliably, and builders keep control over which actions are available.