Connect Webflow to Claude Code
Read a Webflow site's CMS collections and pages, stage item changes, and publish when you mean to. Add Toolspoke to Claude Code once and it can use 11 Webflow actions. Your Webflow credential stays in Toolspoke, and you choose which actions Claude Code may call.
Set it up in Claude Code
Three steps, the first one once for every toolkit you will ever add.
Add Toolspoke to Claude Code
Run it in your project, then /mcp to sign in
claude mcp add --transport http toolspoke https://toolspoke.com/mcpThe sign-in opens in your browser, and Toolspoke asks you to approve it.
Install Webflow and connect it
Add Webflow to a project in your Toolspoke workspace.
API token. You provide it once, when you install the connector. Toolspoke encrypts it at rest and decrypts it only for the length of a single call, and the gateway attaches it to the outbound request itself, so it is never part of the arguments an agent sends.
- API token
- webflow.com → your site → Settings → Apps & integrations → API access → Generate API token. Grant it cms:read, cms:write, pages:read, sites:read and, only if this connection should be able to publish, sites:write. A site token reaches exactly one site, which is the narrowest thing to hand over.
Choose what Claude Code may do
All 11 actions start on. Switch off any of the 3 actions that change something in Webflow, and Claude Code gets a refusal if it tries.
What you can ask Claude Code for
A few of the Webflow actions that only read, so they are safe to try first.
- Read one site: its name, workspace, timezone, custom domains with their ids, and `lastPublished`
What Claude Code can do in Webflow
8 actions read, 2 write and one deletes or permanently changes something.
Reads
List sites
List the sites this token can reach, with each site's id, display name, short name, custom domains, timezone and when it was last published. Free, like every read here. Webflow bills nothing per call and only rate-limits, at 60 requests a minute on Starter and Basic plans and 120 on CMS and above. The `id` values are what every other operation in this connector takes. This is the connection test: it needs no arguments, creates nothing, and a token scoped to the wrong site shows up here as an empty or unexpected list rather than as a puzzling 404 later.
Get site
Read one site: its name, workspace, timezone, custom domains with their ids, and `lastPublished`. Free. Two things here are worth the call, `customDomains[].id`, which is what publish_site takes (it wants domain ids, not domain names), and `lastPublished`, which tells you whether staged CMS edits are already waiting to go out. A site with a `lastPublished` older than its `lastUpdated` has unpublished changes, and publishing will release all of them, not only yours.
List collections
List the CMS collections on a site, blog posts, team members, case studies, whatever this site models, with each collection's id, display name, singular name and slug. Free. Collection ids are what the item operations take. This returns the collections themselves and not their fields; get_collection is where the field definitions are.
Get collection
Read one collection's schema: every field with its `slug`, `type`, `displayName`, whether it `isRequired`, and any validations. Free. Call this before create_item or update_item, always. `fieldData` is keyed by these field slugs and Webflow rejects a key it does not recognise, so a body built by guessing from a field's label rather than its slug fails as a validation error rather than being ignored. `name` and `slug` are required on every collection. Reference fields take another item's id; asset fields take {"fileId":"…","url":"…"}. Structure, not content, which is why this one is logged like any other read.
List collection items
List the items in a CMS collection, the staged copy, which is what the Designer shows and what a future publish would release. Free. Paging is offset and limit, and the response carries `pagination` with the total. Narrow with `name` or `slug` rather than fetching everything: a collection of a thousand blog posts is a very large answer, and each item carries its whole `fieldData`, including rich-text bodies. `isDraft: true` on an item means it will not be published even when the site is; `isArchived: true` means it has been taken out of the collection without being deleted. The payload is somebody's site content, so it is kept out of the audit log.
Get collection item
Read one CMS item by id, its whole `fieldData`, plus `isDraft`, `isArchived`, `lastPublished` and `lastUpdated`. Free. This is the call to make before update_item: a PATCH replaces the fields it names and leaves the rest alone, so reading first is how you know what you are about to change. An item whose `lastUpdated` is newer than its `lastPublished` has staged edits that publish_site would release. The payload is somebody's site content, so it is kept out of the audit log.
List pages
List the static pages of a site with their ids, titles, slugs, parent ids, SEO title and description, Open Graph fields and whether each is a draft or archived. Free. Page ids are what get_page_content takes, and also what publish_site's `pageId` takes when only one page should go out. This returns page settings rather than page copy, the copy is in get_page_content, so it is logged like any other read.
Get page content
Read the text content of one page as a list of DOM nodes, every heading, paragraph and rich-text block the Designer put there, each with its own node id. Free. This is how to audit what a page actually says without opening Webflow, and it is the only read here that returns page copy rather than settings. It reads the staged version, so a page edited but not published shows the new text. Paging is limit and offset over the node list; a long marketing page can be hundreds of nodes. The payload is the site's own prose, so it is kept out of the audit log.
Writes
Create collection item
Add an item to a CMS collection, staged. Nobody outside the workspace sees it until publish_site runs, which is exactly why this is a write and publish_site is destructive. Webflow also offers an /items/live route that skips staging and goes straight to the public site, and this connector deliberately does not declare it. `fieldData` must use the field slugs get_collection returns, and must include `name` and `slug`; an unknown key is rejected rather than ignored. `isDraft` defaults to true on this route, so an item created without it stays a draft even after the site is published, pass false when the item is meant to go out with the next publish. Free to call. The item body is the caller's own prose, so it is kept out of the audit log in both directions.
Update collection item
Change fields on an existing CMS item, staged. Only the fields named in `fieldData` are replaced; anything omitted keeps its current value, so read the item with get_collection_item first if you need to know what that is. Nothing reaches the public site until publish_site runs, the /items/live route that would skip staging is deliberately not declared here. Free to call. This overwrites the previous value of every field it names and Webflow keeps no version history through the API, so the old text is gone: treat it as an edit that cannot be undone from here even though it is not yet public. Setting `isArchived: true` takes the item out of the collection without deleting it, and setting `isDraft: true` holds it back from the next publish.
Deletes or permanently changes
Publish site
Put the site live. This is the operation that makes changes public, and it cannot be quietly undone: Webflow has no unpublish that restores the previous build, and everyone who visits the domain sees the new one from the moment it finishes. It is destructive for a second reason too, it releases every staged change on the site, not only the ones this agent made. Somebody else's half-finished edit in the Designer goes out with yours. Call get_site first and compare `lastUpdated` against `lastPublished` to see whether there is other work waiting. `customDomains` takes domain *ids*, from get_site's `customDomains[].id`, not domain names; `publishToWebflowSubdomain` publishes to the free .webflow.io address and defaults to false. Passing neither publishes nothing, so name at least one target. Needs a token with sites:write; a read-and-write grant that stops short of destructive cannot reach this, which is the point.
Webflow in Claude Code: questions
Does Claude Code see my Webflow credential?
No. The credential is stored encrypted in Toolspoke, and the gateway adds it to each call. Claude Code only ever talks to Toolspoke, and the dashboard shows the last four characters of the credential and nothing more.
Can I stop Claude Code from changing things in Webflow?
Yes. Each of the 11 actions can be switched off for a project. Turn off the 3 that change something and Claude Code can only read.
Will 11 actions crowd Claude Code's context?
No. Toolspoke shows Claude Code five bridge tools, plus a guide where a toolkit has one, and Claude Code searches for the Webflow action it needs when it needs it. Adding more toolkits does not add more tools to its list.
Can I add it once for every project?
Yes. Add --scope user to the command and Claude Code keeps Toolspoke in every project on this machine. Without it, the server is added to the project you ran the command in.
What does it cost?
Tool calls are free on every plan, and a Webflow call does not run in a sandbox, so it bills nothing.
Webflow in your other apps
Try Webflow in Claude Code
Free to start, and tool calls cost nothing on any plan.