DeepSeek Harness Controls Figma via Open-Source Plugin for UI Design & Animations

Open-Source Plugin Enables DeepSeek Harness to Operate Figma

Developer Douasa recently shared on X that the DSH-to-Figma plugin has been updated to version 0.3.5, unexpectedly receiving a like from Cui Tianyi. This plugin allows DeepSeek Harness (DSH) to directly control the Figma design canvas.

Installation and Setup

To use the plugin, first download the Figma desktop app . Then, input the following prompt into DSH to automatically install or upgrade the plugin:

帮我安装或升级 DSH to Figma：
Duoasa/DSH-to-Figma
按仓库的 docs/setup.md，使用当前宿主的 CLI、profile 和 DSH_HOME，
安装 dsh-to-figma@latest，保留现有配置，使用 local 模式。
安装后调用 figma_setup，给我本机 manifest.json 的路径并指导导入 Figma。
若已安装，更新原目录并提醒我重新运行 Figma 插件，最后用 figma_status 确认连接。

When a connection prompt appears in the Figma canvas showing a connected status, DeepSeek Harness has successfully integrated with Figma. Important: Do not close this prompt window; it must remain open for DSH to continue reading and modifying Figma designs.

01 DSH Takes Over Figma Canvas: UI Design and Animations

When creating interfaces, users can instruct DeepSeek Harness to generate pages, frames, text, and vector graphics directly in Figma based on a textual description, organizing elements via auto-layout.

If a reference screenshot exists, DSH can also reconstruct the interface from the screenshot . The author tested this by providing a screenshot of the DeepSeek web interface.

Upon receiving the task, DSH first analyzed the target frame's dimensions, existing layers, and design content, then asked which replication precision to use. Three generation strategies are offered:

Balanced replication – a middle ground between speed and detail.

Quick layout – faster, less detailed.

Fine replication – high fidelity, more time-consuming.

The author chose balanced replication. DSH then generated the corresponding page in Figma.

The generated content consists of multiple independent design elements. Users can inspect the page structure in the left layers panel, select individual text, graphics, and frames for further adjustments.

DSH-to-Figma provides screenshot and visual verification capabilities . It automatically captures the Figma canvas and returns the screenshot via DSH's native image attachment mechanism, allowing DSH to verify layout and modifications against the actual visual output.

Beyond UI design, DSH-to-Figma supports Figma's native animation capabilities , including creating and adjusting keyframes, text animations, animation styles, and timeline information. It can also read existing animations and convert some animation properties to CSS.

The author asked DSH to add an animation to the generated page.

The resulting animation appeared natural. After generation, users can continue editing the animation in Figma's native timeline, enabling a seamless creative workflow.

Additionally, the plugin supports exporting animations in MP4, GIF, and WebM formats via the editor's native interfaces.

02 Local Bridge + Native API: Connecting DSH to the Design Canvas

DSH-to-Figma registers a series of Figma operation tools with DeepSeek Harness via the DSH plugin mechanism, enabling the model to invoke corresponding functions based on user instructions.

When the model initiates a tool call, DSH-to-Figma passes the operation request through a local bridge service to the companion Figma plugin running in the Figma desktop app. That plugin then calls Figma's public Plugin API to perform canvas reading, layer creation, property modification, and asset export.

On first installation, DSH automatically generates a paired Figma companion plugin for the current machine. Users only need to import the manifest file once in Figma and launch the companion plugin to establish the connection.

DSH-to-Figma also includes a built-in figma-workbench skill that provides the model with specialized guides for canvas operations, design-to-code conversion, component and variable management, helping the model better understand and execute design tasks.

The project defaults to Local mode , which does not rely on the official Figma MCP for canvas operations, thus avoiding consumption of official MCP call quotas.