Webflow hosts an MCP server at mcp.webflow.com. It reaches the CMS and site structure rather than the visual canvas — which is the right half, because bulk content work is exactly what the designer is worst at.
https://mcp.webflow.com/mcp
Claude Sonnet 4.5
Chat with 60+ AI models on the same workflow — switch to a different model mid-conversation and re-run the same prompt, or use Compare mode to put several side-by-side and balance quality vs. cost.
OAuth against your Webflow account. Scope the authorised site list deliberately — the agent can write to any site you grant.
How models use it and what it is built for.
The server exposes Webflow’s Data API through MCP. An agent can list sites, read page metadata, and — most usefully — work with CMS collections: enumerate collections and their field schemas, read items, and create or update them. That makes systematic content work tractable. Retagging two hundred blog entries, filling missing SEO descriptions, or auditing which items reference a category that no longer exists are all single questions here and afternoon-long clicking exercises in the designer. Publishing is controllable, so changes can be staged rather than pushed live immediately. What it does not do is design: layouts, styles and interactions remain the designer’s domain, and an agent cannot restructure a page visually. Treat it as a content and metadata surface, and it is genuinely useful.
Typical tools an AI model can call. Exact names vary by version.
Copy any of these into MCP Agent Studio after connecting.
Which blog collection items are missing a meta description or an og image?
Update the author reference on every post tagged "engineering" to the new team entry.
List collection items that reference a category which no longer exists.
Draft and create a case study item following the structure of the three most recent ones.
This is not a single-model product: you get the same MCP connection with 60+ models (Claude, GPT, Gemini, DeepSeek, open-weight, and more), you can switch mid-conversation, and you can open Compare mode to run the same prompt against multiple models at once. The card above is a suggested starting point for this server — not the only choice.
Default pick for Webflow
Claude Sonnet 4.5
Collection items are nested objects with reference fields that break easily. Sonnet 4.5 respects the field schema on writes instead of flattening it into plain strings.
Listing tools proves the server is reachable, not that a model can work with it. Evals go further: they read every tool on the server, write a test suite from its real schemas, and run it — code decides pass/fail on the responses (schema conformance, error codes, pagination, result caps) while a scoring model grades plain-English tasks driven through the tools.
Get a pass/fail report per tool with the evidence behind each verdict — and replay the same suite after every schema change. Destructive tools are excluded from the run.
Run evalsOpen MCP Agent Studio with the connection pre-filled. Add your token, pick any of 60+ models, and start chatting — no install required.
Open Agent StudioCommon questions about connecting, scoping and using it safely.
Webflow’s official hosted MCP server at mcp.webflow.com. It exposes sites, pages and CMS collection items as MCP tools, so an AI assistant can read and edit structured content programmatically.
No. It reaches the Data API — content, CMS items and metadata — not the visual canvas. Layout, styling and interactions stay in the designer, which is a sensible boundary given how much damage an agent could do to a live layout.
Not unless you publish. Changes can be staged and published as a separate step, so the safe pattern is to let an agent make the edits and keep publishing as a human decision.
That is arguably its best use. Finding every item missing a meta description and filling it correctly is mechanical, high-volume and error-prone by hand — and it is the kind of task that never gets scheduled because it is so tedious.
Webflow couples content to a visual site builder, so it suits marketing sites where the same team designs and writes. Sanity is a headless content backend with a stricter schema and a real query language, which gives a model more to reason over.