collections_items_create_item and collections_items_update_items write drafts, the live variants push straight to the published site, and collections_items_publish_items promotes drafts by id — so an agent can be given the draft path and nothing else. The Designer tools only work while a Designer session is open and connected, and element creation caps at 3 levels in one call, so a deep structure is built across several passes. The CMS side assumes you are already holding site and collection ids, which makes the two listing tools the first calls in almost every sequence.
Webflow's own server over its Data API plus a bridge into an open Designer session — CMS collections and items, pages, components, scripts, sites and canvas editing — across 43 tools.
- CMS schema work: collections created and inspected with their field definitions, and static, option and reference fields added or updated on an existing collection
- Items written either way round — created and updated as drafts and then published by id, or created and updated directly against the live site
- Page settings including SEO metadata, Open Graph data, slug and publish status, read and written per locale
- Localisation kept on its own path, with text nodes and property overrides updated for secondary locales on both static pages and components
- Site publishing to named custom domains or the Webflow subdomain, with registered and applied scripts listed and inline scripts registered up to 2,000 characters
- Designer-side building on the active page — elements no more than 3 levels deep per call, plus styles, variables, component instances, assets and the current selection
WEBFLOW_TOKEN for an account that holds the sites you name, and the site and collection ids the listing tools return. The Designer half additionally needs the Webflow MCP app connected to an open Designer session, which get_designer_app_connection_info reports on. MIT.
One command plus a key — npx -y webflow-mcp-server@latest, then supply credentials