Your website just became something any capable agent can operate. Your job is the system it works within and the review before it ships.
What Framer MCP is and how designers should use it
Framer MCP is a remote Model Context Protocol server, released on 22 October 2026, that lets an outside AI tool such as Claude, Claude Code, ChatGPT, Codex or Cursor work directly inside your Framer project. According to Framer's MCP page, once you approve the connection the agent can work with the canvas, components, supported project settings and the CMS, and it only publishes when you explicitly ask it to.
The short answer for designers: use Framer MCP for the repetitive, multi-step site work you already know how to judge (CMS entry, page scaffolding, content updates, cross-tool chores), keep every change on a branch, and stay the person who decides what ships. It is not a replacement for taste. It is a faster pair of hands that happens to live in the chat tool you already use.
The official announcement frames it as connecting "AI tools to Framer". That sounds small. The practical shift is bigger: your site stops being something you only edit inside Framer and becomes something any capable agent can operate, alongside your files, your Figma references and your inbox.
What changed on 22 October
Before this release, connecting an outside agent to Framer meant the older External Agents setup, a command line tool. Framer now calls MCP "the recommended way to connect agents", and the page lists several practical differences:
There is no local Framer bridge to install, and the Framer editor does not need to stay open.
Connection happens over OAuth, and the agent uses your account's permissions, so it only reaches projects your Framer account can already access.
By default, changes stay in your project and do not touch the live site.
AI usage follows the plan or credits of the agent you connect, not Framer's built-in agent.
Framer is also honest about a gap: the older External Agents setup still covers some workflows, localization among them, that Framer MCP does not yet cover. Both stay supported while MCP reaches feature parity.
How the three ways of using AI in Framer compare
Criteria | Built-in Framer agent | Framer MCP (new) | External Agents CLI (older) |
|---|---|---|---|
Where you work | Inside the Framer editor | In your own AI tool (Claude, Codex, Cursor and others) | In a terminal |
Setup | None | Add the server or plugin, approve with OAuth | Install the CLI |
Editor must be open | Yes | No, per Framer | Depends on the workflow |
Context it can combine | Your Framer project | Your project plus the agent's own files, connectors and tools | Your project plus local files |
Who pays for AI usage | Framer AI credits | The connected agent's plan or credits | The connected agent |
Localization workflows | Supported in editor | Not yet, per Framer | Supported |
Best for | Quick visual edits while designing | Multi-step jobs that span tools | Existing scripted pipelines |
The verdict: keep the built-in agent for in-the-moment design edits. Pick Framer MCP when the job pulls information from somewhere else, such as a folder of case studies, a brief in Google Drive or references in Figma. Stay on the CLI only if you depend on localization or an existing script.
How to set up Framer MCP in ten minutes
The Framer Academy lesson on MCP walks through the setup. In Codex, open the Plugins tab, search for Framer, add the official Framer plugin and allow access to your account. In Claude or Claude Code, add the Framer MCP server and approve the connection with your Framer account. For any other client that supports remote MCP, Framer publishes a manual guide and the server address https://mcp.framer.com/mcp.
Then point the agent at a project. Open the project in Framer, click Invite, copy the Project Link, start a new chat, type "Connect to" and paste the link. In a Codex project you can ask the agent to remember the site so later chats skip this step.
One setup habit pays off immediately: before the first real task, ask the agent to describe the project back to you. Page list, CMS collections, component names. If its summary is wrong, its edits will be wrong too, and it is cheaper to find out in a read-only question.
The Branch, Review, Publish loop
The most useful line on Framer's page is the quiet one about branches. You can ask the agent to work on a branch, review its changes, compare them with main, and publish when ready. Build every MCP session around that.
1. Brief on a branch. Start each job with "create a branch called X and work only there". Give the goal, the pages in scope and what must not change (navigation, footer, global styles).
2. Constrain the job. Agents drift when the task is vague. "Add the six case studies from this folder to the Work collection, using the existing fields, no new fields" is a good brief. "Improve the work section" is not.
3. Review like a design lead. Open the branch in Framer and check it visually at desktop, tablet and phone widths. Read every new CMS entry. Look for invented copy, wrong image crops and broken links. Compare against main.
4. Publish yourself, or say the word. Framer states the agent publishes only when explicitly asked. Make that request the last line of the job, after your review, never part of the opening brief.
If you already write reusable instructions for the built-in agent, see the guide to Framer Skills for designers. Design system rules written there are the same rules you want an outside agent to respect.
Five jobs worth handing to Framer MCP this week
Framer's own examples point at the real value, which is crossing tools:
Folder to CMS. Sort a local folder of case studies, articles and photos and prepare them as CMS entries.
Figma to page. Pull design references from Figma and scaffold a new landing page on a branch. Pair this with the Figma MCP server workflow for designers so both ends speak through the same agent.
Brief to draft. Read a client brief from Google Drive and draft page copy into the CMS for review.
Feedback to fixes. Collect feedback from Slack and turn it into a checklist of edits on a branch.
Scheduled checks. Framer mentions scheduled tasks such as weekly keyword checks or analytics reports. Good for maintenance, less good for anything visual.
Where it struggles is the same place every agent struggles: judgment. Spacing rhythm, hierarchy, motion feel and whether a headline actually lands are still yours. Treat any visual change it proposes as a sketch, not a decision.
Hype vs reality
The hype says you will run your whole website from a chat window. Partly true. Content operations, CMS work and page scaffolding genuinely move faster when the agent can read your files and write to Framer in one session.
What does not hold up: the idea that it designs for you. The agent works with the components and styles you already built. A weak design system gives you fast, consistent, weak pages. Framer also lists gaps, localization for one, and the announcement does not describe plan limits or pricing, so check those before you build a client workflow on it.
Verdict: worth adopting now for content and structure work, with the Branch, Review, Publish loop as a non-negotiable habit.
Where this takes web design next
Tools like Framer MCP turn a website into something agents can operate, the same direction covered in the DIGCY piece on the future of landing pages. Over the next one to three years, expect the designer's job on a site to tilt toward three things: building the system the agent works within (components, styles, CMS structure), writing the rules it follows, and reviewing what it produces. The person who can brief, constrain and review an agent well will ship more than the person who only pushes pixels, and more than the person who lets the agent run unchecked.
Frequently asked questions
Will Framer MCP replace web designers?
No. It replaces some of the hours designers spend on CMS entry, copy updates and repetitive page building. It cannot decide what the site should say, how it should feel or whether a layout works. If your role is mostly manual content entry, that part is shrinking, so move toward owning the system, the brief and the review.
Is Framer MCP overhyped?
Partly. The cross-tool workflows are real and useful today. The claim that it designs sites for you is not what Framer says it does: it works on your canvas, components, settings and CMS, and its output is only as good as the system you built.
Which AI tools work with Framer MCP?
Framer's page lists ChatGPT, Codex, Claude, Claude Code, Cursor, VS Code, Windsurf, Notion and others, and says any tool that supports remote MCP servers and OAuth can connect.
Can Framer MCP publish my site without asking?
According to Framer, changes stay in your project by default and the agent publishes only when you explicitly ask. Working on a branch adds a second layer of safety.
Does Framer MCP use my Framer AI credits?
Framer states that AI usage follows the plan or credits of the agent you connect, not Framer's built-in agent.
Am I falling behind if I have not tried it yet?
Not if you start this week with something small. Connect it to a test project, run one CMS task on a branch and review it. That single loop teaches most of what matters.
Getting help with the system behind it
Framer MCP rewards teams with a clean component library, sensible CMS structure and clear content rules. If your site needs that groundwork first, a website redesign or focused web app design work is where to start, and an experienced UI/UX agency in Casablanca can set up the system so both people and agents build on it well. DIGCY is a Dribbble Selected Agency working with clients globally, and products that go beyond a website fit AI product development.
Let’s keep in touch.
Discover more about high-performance web design. Follow us on Twitter and Instagram.
