Webflow

Implements Model Context Protocol as a translation layer between AI agents (Cursor, Claude Desktop) and Webflow's REST API, supporting dual deployment modes: Node.js with stdio communication for local development and Cloudflare Workers with Durable Objects for stateful cloud execution. The server exposes Webflow resources (sites, pages, CMS collections) as MCP tools with schema-based function definitions, enabling AI agents to discover and invoke operations through a standardized interface rather than direct HTTP calls.

Exposes MCP tools to list all Webflow sites accessible to an authenticated user and retrieve detailed metadata (site ID, name, domain, publish status, last modified timestamp) for individual sites. Implements pagination and filtering through Webflow's REST API, tracking publish state to enable agents to determine which sites have pending changes requiring deployment.

Provides MCP tools to list pages within a site, retrieve page metadata (title, slug, SEO settings, custom attributes), fetch page content (HTML/DOM structure), and update page settings and content. The implementation maintains awareness of page hierarchy (parent-child relationships) and supports bulk operations on multiple pages through sequential tool invocations, enabling agents to restructure site navigation or update content across page trees.

Exposes MCP tools to list CMS collections within a site, define collection fields with type constraints (text, number, date, reference, multi-reference), and perform CRUD operations on collection items. The implementation validates item data against field schemas before submission to Webflow API, preventing invalid data from reaching the server. Supports reference fields (linking items across collections) and multi-reference fields (one-to-many relationships), enabling agents to build and maintain relational data structures.

Provides MCP tool to publish pending changes from a Webflow site to its live domain. The implementation tracks which resources (pages, CMS items) have unpublished changes and enables agents to trigger deployment atomically, publishing all pending changes in a single operation. Supports conditional publishing (only if changes exist) to avoid unnecessary API calls and deployment cycles.

Implements Webflow API token authentication at the MCP server level, validating tokens and enforcing scope-based access control for all tool invocations. The server stores the API token securely (environment variable or Cloudflare Workers secret) and includes it in all outbound Webflow API requests. Scope validation ensures that tools attempting to write data (pages:write, collections:write) are only available if the token has the required permissions, preventing agents from attempting operations that will fail.

Abstracts deployment environment differences through a unified MCP server implementation that runs in two modes: Node.js with stdio transport for local development (connecting to Cursor/Claude Desktop via process pipes) and Cloudflare Workers with Durable Objects for cloud deployment (connecting via HTTP). The abstraction layer handles transport-specific concerns (stdio buffering, HTTP request/response serialization) while exposing identical MCP tool interfaces in both environments, enabling agents to switch deployment modes without code changes.

Automatically generates MCP tool schemas for all Webflow operations (list sites, update page, create collection item, etc.), exposing them through the MCP tools list endpoint. Each schema includes parameter definitions with types, descriptions, and required/optional flags, enabling MCP clients to discover available operations and validate parameters before invocation. The server validates incoming tool calls against schemas, rejecting malformed requests with detailed error messages before forwarding to Webflow API.