# Building the Engine While Flying the Plane: Launching the Figma MCP Server — Jesse Lumarie, Figma **URL:** \ **Category:** Content Creators\ **Tags:** mcp, ai-engineer\ **Created:** [28 August 2026 14:00 UTC](https://www.artofsm.art/t/building-the-engine-while-flying-the-plane-launching-the-figma-mcp-server-jesse-lumarie-figma/23305 "2026-08-28T14:00:06Z")\ **Posts on this page:** 2\ **Page:** 1
**Author:** ![merefield](https://www.artofsm.art/user_avatar/www.artofsm.art/merefield/32/74_2.png) [@merefield](https://www.artofsm.art/u/merefield)\ **Post date:** [28 August 2026 14:00 UTC](https://www.artofsm.art/t/building-the-engine-while-flying-the-plane-launching-the-figma-mcp-server-jesse-lumarie-figma/23305/1 "2026-08-28T14:00:06Z")
[https://www.youtube.com/watch?v=ZIYYsAzaLlA](https://www.youtube.com/watch?v=ZIYYsAzaLlA) Jesse Lumarie detailed the rapid development of Figma’s first Multi-Context Protocol (MCP) server, which enables seamless integration of design and code contexts to enhance AI-driven workflows, initially launching as a local developer tool before expanding to a remote server with read/write capabilities. Despite challenges with evolving MCP client support, the project leveraged Figma’s internal systems and a culture of engineering experimentation to create a fast-growing product that advances AI integration within design and development. ---
**Author:** ![merefield](https://www.artofsm.art/user_avatar/www.artofsm.art/merefield/32/74_2.png) [@merefield](https://www.artofsm.art/u/merefield)\ **Post date:** [28 August 2026 14:24 UTC](https://www.artofsm.art/t/building-the-engine-while-flying-the-plane-launching-the-figma-mcp-server-jesse-lumarie-figma/23305/3 "2026-08-28T14:24:03Z")
Jesse Lumarie, a software engineer at Figma, shared the journey of building Figma’s first MCP (Multi-Context Protocol) server within three months. The MCP server facilitates seamless context sharing between production code and design, enabling AI tools to integrate with Figma without dedicated integrations. The project began in late 2024 after Anthropic released the MCP server specification, which was initially adopted by few AI clients. Jesse started developing a Figma plugin-based MCP server as a side project, eventually forming a team known as the “MC Peeps” to advance the effort despite evolving and sometimes inconsistent client support for the MCP spec. The initial launch focused on a local MCP server tailored for developers, allowing them to pull detailed design context such as component data, spacing, and variables into their coding environments. Figma’s internal scene graph, represented in C++, was converted into various formats for the server, including an abstract JSX/XML-like representation, a React Tailwind format, and images. The React Tailwind format was favored for its fidelity and compatibility with AI models, while images were used as supplementary context. To evaluate the quality of outputs, the team developed automated evaluation tools, moving away from tedious manual grading to more scalable, AI-assisted assessments. A critical enhancement involved integrating Figma’s existing “code connect” system, which links design components to corresponding codebase components. This allowed the MCP server to produce outputs referencing battle-tested, accessible, and internationalized components rather than generic HTML, improving both fidelity and efficiency by reducing context size. Despite challenges with incomplete client support for advanced MCP features like server instructions, elicitation (user input queries), and sampling (server querying client LLMs), the team implemented creative workarounds to enhance user experience and component mapping. Security, file permission respect, and pricing considerations were paramount in deciding the architecture. Initially, the MCP server was local, leveraging Figma’s Electron-based desktop app to maintain data privacy and accelerate development. After internal launch and positive community feedback, Figma released a remote MCP server in late 2025, supporting both read and write capabilities. This innovation contributed to one of Figma’s fastest-growing products, demonstrating the power of combining design and code contexts through AI-driven workflows. Jesse concluded by emphasizing how early the MCP ecosystem still is, with the spec only two years old and many best practices yet to be established. He praised Figma’s culture of empowering engineers to experiment and build new products, even without formal staffing, which led to further innovations like “make,” Figma’s agent solution for working with local codebases and GitHub. Jesse encouraged attendees to engage with him for further discussions on MCP and Figma’s evolving AI integrations. ## Useful Links - [Anthropic MCP Server Specification](https://github.com/modelcontextprotocol) — Directly substantiates the protocol Figma implemented. - [Figma Code Connect Documentation](https://www.reddit.com/r/FigmaDesign/comments/1qfbuw5/figma_code_connect_is_it_worth_it/) — Explains the component linking system used to improve MCP server output fidelity.