How AI generated and managed content works on a Webflow site, why Webflow's built-in AI credits aren't required, and the issues you'll run into when an AI connector does the building.
Webflow's MCP server connects an AI agent directly to a Webflow site. Claude and ChatGPT have official Webflow connectors, and Gemini can connect through the MCP server URL. From there the agent can generate pages and designs, abstract them into components and elements, and build out a full page.
Describe the page, section, or change.
Claude, ChatGPT, or Gemini plans the build.
Creates pages, styles, components, and elements.
Native Webflow content you can edit in the Designer.
This site was built with Claude through the Webflow connector: the pages, SEO settings, styles, the shared nav and footer components, and every section you're reading.
Webflow's built-in AI engine runs on AI credits from your Workspace plan, 200 to 400 per month on the standard plans. You don't need any of them to build with an AI connector.
Issues you'll likely run into when an AI connector is doing the work.
For most of 2026 the MCP server couldn't create Webflow Interactions at all; IX3 support was only added in a fall 2026 update, and it's the least reliable part of the toolset. Webflow's own guidance warns that some interaction definitions save without errors and then never animate, and the agent can't tell the difference. Plan to build or verify every animation by hand in the Designer or on a published page.
The connector is limited to connecting one Claude user or account to one Webflow site. There is currently no switching between multiple sites, which gets in the way fast if you manage more than one project.
The Webflow MCP Bridge App is a Webflow Designer app that links the agent to your live Designer session. Most building no longer needs it, but anything visual does: element snapshots, plus reading or changing the current selection, page, canvas, and breakpoints. It only works while the Designer is open in a browser tab with the Bridge App running, and that tab has to stay active in the foreground. Idle or background tabs time out, and the visual tools only see whichever page the canvas has open. The AI's eyes depend on you keeping a specific tab open on a specific page, which is limiting when you want it to work on its own or across several pages.