Lovable: Build in a Draft from Your AI Client
AI clients connected to the Lovable MCP server, such as ChatGPT and Claude, can now create a draft of a project and build in it. Code changes stay in the draft until the user accepts it into the project, either in Lovable or by asking the client. Building in a draft counts as standard build usage and consumes credits.
Key Takeaways
- MCP-connected clients like ChatGPT and Claude can now create a draft and build in it.
- Changes remain isolated in the draft until the user accepts them into the project.
- Acceptance works in Lovable or by asking the AI client.
- Drafts add a review checkpoint before external AI edits reach the main project.
- Building in a draft is standard build usage and consumes credits.
- The update makes the Lovable MCP server a safer route for building from outside the Lovable editor.
Drafts for MCP-connected AI clients
Lovable updated its MCP server so that connected AI clients, such as ChatGPT and Claude, can create a draft of a project and build inside it. Previously, an external assistant working on a project had no isolated space for its changes; drafts provide one.
How it works
When a client builds in a draft, code changes stay in the draft and do not touch the main project. They only land when the user accepts the draft into the project. Acceptance can be done in Lovable itself, or by asking the AI client to do it. This mirrors how drafts work in the Lovable editor and gives users a review step before external AI changes go live.
Usage and credits
Building in a draft is treated as standard build usage and consumes credits, the same as building in the editor.
Why it matters
Letting a general-purpose assistant edit an app can feel risky. Drafts reduce that risk by keeping experiments separate until approved, which makes it more comfortable to delegate real work to ChatGPT or Claude while keeping Lovable as the place where the project is reviewed and accepted.