Figma MCP connects to Weave: your custom AI tools land in ChatGPT and Claude

What Figma just announced

Since August 11, 2026, Figma has let you run its Weave tools directly from external AI clients like ChatGPT, Claude, or Cursor, thanks to its MCP (Model Context Protocol) server. In practice: you describe what you need to your favorite agent, it finds the matching Weave tool, runs it, and sends the result back to you without ever leaving your conversation.

A quick refresher for anyone who missed it: Figma Weave is Figma's building block for generative AI workflows (images, videos, brand assets) built on a visual canvas. MCP, on the other hand, is the protocol that lets AI agents talk to external tools in a standardized way. By connecting the two, Figma opens up its Weave workflows to the entire ecosystem of MCP-compatible agents.

How it actually works

Take the example Figma gave: you have built a Weave tool that turns product photos into brand mockups. Instead of opening Figma, navigating to Weave, and running the tool manually, you simply ask your agent: "Run my brand mockup tool in Figma Weave with these images." The agent identifies the tool, runs it, tracks its progress, and delivers the result right in the chat thread.

• You stay in your workflow, without switching apps
• Context is preserved: prompts, images, brand guidelines
• You keep chatting with your agent while results come back

Figma has already teased what is next: soon, you will be able to build new Weave workflows without ever leaving your agent. The direction is clear: Figma wants Weave to become an execution engine you can access from anywhere, not just from its own canvas.

Why this fits a bigger trend

This release is not an isolated move. For several months now, Figma has been building bridges between its ecosystem and external coding agents: Codex, Claude Code, and GitHub Copilot have all gotten their own dedicated MCP integration. The direction is crystal clear: Figma no longer wants to be just a destination, but a capability you can call from wherever the actual work happens — often a terminal or an AI chat window.

Our take at Mazette.co

At Mazette.co, we use Figma every day to design interfaces before translating them into Webflow. This announcement speaks to us directly, and for a simple reason: it shrinks the distance between an idea and the final asset even further.

On paper, it is a good idea, almost an obvious one. Teams already using AI agents to code, research, or organize their work clearly do not want to reopen ten tabs just to generate a brand image or a product mockup. Being able to trigger a Weave tool from Claude or ChatGPT, without losing the thread of the conversation, fits perfectly with how we increasingly work: as a continuous flow, not as back-and-forth trips between siloed apps.

What strikes us as particularly smart is the context preservation. In an agency setting, we constantly juggle brand guidelines, reference visuals, and client instructions. Being able to keep that context from start to finish, without retyping it at every step, is a real time-saver — and it cuts down on the risk of mistakes.

Our small caveat: this feature assumes your Weave tools are already well built and documented beforehand. An external agent only finds and runs an existing tool — it does not replace the work of designing the workflow itself. For an agency like ours, that means investing time upstream to make sure these Weave tools are genuinely usable by third parties, whether human or agent.

Fundamentally, this echoes the same philosophy behind Figma Motion and the agent-generated shaders unveiled at Config 2026: Figma is working to make its creative building blocks accessible through conversation rather than manual navigation through an interface. It is a coherent direction, and honestly, a pretty exciting one for teams that already know how to make the most of AI agents in their day-to-day work.