Sketch

Sketch offers a downloadable agent skill called 'sketch-implement-design' that instructs compatible AI coding agents to convert Sketch design frames into working frontend code by following a defined inspection and extraction workflow.

The skill, published in Sketch's open GitHub skills repository, guides AI agents such as Claude Code and Cursor through a step-by-step process: inspecting the design, extracting visual hierarchy, gathering colors and typography, exporting needed assets, and writing the code implementation. It works in conjunction with the Sketch MCP server and supports frame-level targeting via Sketch document URLs. The skill includes guardrails for large documents and correct asset export. It is available free of charge to any user with a compatible AI client.

Sketch integrated a built-in MCP (Model Context Protocol) server into its Mac app, allowing users to connect external AI coding agents — such as Claude Code, Cursor, or Codex — directly to their open design documents to automate tasks like exporting assets, renaming layers, swapping symbols, and generating placeholder content.

Released in October 2025 with Sketch version 2025.2.4, the MCP server exposes two core tools: get_selection_as_image, which captures a screenshot of the user's current canvas selection and returns it to the AI client for analysis, and run_code, which lets the AI write and execute SketchAPI scripts to manipulate design files programmatically. The server is local-only, off by default, and cannot be accessed remotely — all processing happens on the user's machine. Users must actively start the server from the Command Bar. Sketch also published a skills repository, including the 'sketch-implement-design' skill, which guides compatible AI agents through a structured workflow to convert Sketch frames into working frontend code.

Sketch deployed a one-click image background removal feature in its Mac app, using Apple's on-device machine learning to automatically remove backgrounds from images placed in design documents — with no data sent to external servers.

Released in November 2025 as part of the 'Copenhagen' update (Sketch 2025.3), the feature uses Apple's Vision framework to run machine learning models locally on the user's Mac. It offers two modes: one optimized for images containing people, and another for objects. Because processing happens entirely on-device, no image data leaves the user's computer. The feature is accessible via the JavaScript plugin API as well, allowing plugin developers to call it programmatically on single or multiple image layers.