OpenPencil

Design teams that depend on Figma face a familiar set of tradeoffs: a closed file format, browser-locked editing, and no real way to script the canvas or hand it to an automated agent. OpenPencil is an attempt to answer those frustrations with an open-source design editor that can open native .fig files directly, rather than forcing an import or conversion step. It ships as both a desktop application built with Tauri and a web PWA, and it positions itself as a toolkit rather than just an app: a programmable surface for design rather than a sealed product. The pitch is straightforward. Keep the Figma files you already have, edit them in software you control, and wire the canvas into the same AI and agent tooling you use everywhere else in your stack.

Technically OpenPencil renders with Skia through CanvasKit compiled to WebAssembly, and handles flex and grid layout with a forked build of Yoga. Figma compatibility comes from reading the .fig format directly, which uses Kiwi binary encoding with Zstd compression inside a ZIP container, and the project claims round-trip fidelity plus copy-paste between Figma and OpenPencil. What sets it apart is the agent surface. It exposes an MCP server that works with Claude Code, Cursor, and Windsurf, ships a headless CLI with XPath-style queries for inspecting and exporting files, and offers roughly 90 to 100 design tools through a built-in chat. Real-time collaboration runs peer-to-peer over WebRTC using Yjs CRDTs, so syncing happens without a central server.

The natural audience is designers and developers who want programmatic control over design work and a self-hostable, account-free editor. Concrete uses follow from the feature set. A front-end developer can point a coding agent at a .fig file through the MCP server and have it read components, extract design tokens, and emit JSX with Tailwind classes, closing part of the design-to-code gap. A design-systems maintainer can script bulk edits or audits across files via the CLI instead of clicking through them. Small teams that dislike SaaS lock-in get local-first editing with optional peer-to-peer collaboration. It also suits anyone simply curious about how far AI-assisted, chat-driven design manipulation can go before it stops being a novelty.

The caveats are significant and the maintainers state the most important one plainly: the project is in active development and not ready for production use. That alone should keep it out of any client deliverable or critical workflow for now. The Skia and WASM rendering approach brings a large bundle and harder debugging when something misbehaves on the canvas. Round-trip fidelity with .fig files is a hard problem given Figma's proprietary and evolving format, so edge cases and silent data loss are real risks worth testing before trusting it with originals. The project carries small-organization backing, which raises the chance of slowed momentum or abandonment. A comprehensive test suite is reassuring and unusual at this stage, but tests do not substitute for production hardening.

Measured against Figma, OpenPencil is not a replacement today, and against Penpot, the established open-source option, it is younger and rougher but more ambitious about AI and agent integration. Its real value right now is exploratory: a way to experiment with agent-driven design, scriptable file manipulation, and a Figma-compatible editor you can actually self-host. Adopt it for prototypes, internal tooling, design-to-code experiments, and learning, where breakage is cheap and the agent hooks pay off. Avoid it for anything a client or your team depends on, since the maintainers themselves warn against that. The sensible posture is to watch it closely, contribute if the direction appeals to you, and revisit once a stable release and a broader contributor base signal it has matured.