Website builders are missing an obvious opportunity. As AI agents become standard development tools, platforms like Squarespace, Wix, and Webflow should ship MCP servers.

The Problem

Today, if an AI agent needs to update a Squarespace site, it has to:

  1. Reverse-engineer the admin panel
  2. Scrape authentication flows
  3. Replay network requests
  4. Hope nothing changes

This is fragile, unauthorized, and wasteful. The platform has a perfectly good API internally—they just don't expose it.

What an MCP Server Enables

Imagine a Squarespace MCP server with tools like:

{
  "tools": [
    {
      "name": "create_page",
      "description": "Create a new page",
      "inputSchema": {
        "type": "object",
        "properties": {
          "title": {"type": "string"},
          "slug": {"type": "string"},
          "content": {"type": "string"},
          "template": {"type": "string"}
        }
      }
    },
    {
      "name": "update_page",
      "description": "Update existing page content"
    },
    {
      "name": "upload_image",
      "description": "Upload image to media library"
    },
    {
      "name": "publish_blog_post",
      "description": "Create and publish a blog post"
    }
  ]
}

An AI agent could then:

  • "Add a new portfolio item with these images"
  • "Update the About page to mention our new service"
  • "Create a blog post from this draft"
  • "Change the hero text on the homepage"

No scraping. No reverse engineering. Just structured tool calls.

Why This Makes Business Sense

Stickiness: Users with AI workflows built on your MCP server won't switch platforms easily.

Differentiation: "Works with Claude, ChatGPT, and Copilot" is a feature.

Reduced support burden: AI agents that use official tools don't break when you update your UI.

New market: Agencies and developers building AI-powered content workflows.

The Authentication Model

MCP servers support OAuth and API keys. The flow:

  1. User authorizes the MCP server in Squarespace settings
  2. MCP server receives scoped API token
  3. AI agents connect via MCP protocol
  4. All actions are logged and auditable

Same permissions model as existing integrations, just with a standardized protocol.

What Tools Would Be Essential

Content Management:

  • Create/update/delete pages
  • Publish/unpublish content
  • Manage blog posts
  • Handle media uploads

Site Structure:

  • List pages and navigation
  • Read current content
  • Get site settings

E-commerce (if applicable):

  • Manage products
  • Update inventory
  • Read orders (read-only)

Forms and Data:

  • Read form submissions
  • Export contact lists

The Read-Only Start

Platforms nervous about AI agents writing to their systems could start read-only:

{
  "tools": [
    {
      "name": "list_pages",
      "annotations": {"readOnly": true}
    },
    {
      "name": "get_page_content",
      "annotations": {"readOnly": true}
    },
    {
      "name": "list_blog_posts",
      "annotations": {"readOnly": true}
    }
  ]
}

This alone is valuable—AI agents can understand the site structure and content before suggesting changes through the normal UI.

Who Should Build This First

Squarespace: Has developer-friendly DNA, already has some API capabilities.

Webflow: Developer-focused, already has a robust API.

Wix: Huge market share, would benefit most from differentiation.

Ghost: Open source, could ship this quickly with community support.

WordPress: Already has REST API, just needs MCP wrapper.

The Standard Exists

MCP is an open standard from Anthropic. The specification is public. Reference implementations exist in TypeScript and Python.

Building an MCP server for an existing API is straightforward—it's essentially a translation layer with JSON-RPC transport.

Call to Action

If you work at a website builder:

  1. Look at your internal APIs
  2. Identify which could be exposed as MCP tools
  3. Start with read-only operations
  4. Ship it as a beta feature

The first major website builder with an MCP server will capture the AI-native workflow market. Everyone else will be playing catch-up.

AI agents are becoming standard tools. Give them a proper interface.

React to this post: