Creative Coding

ShaderVine Is a WebGPU Shader Editor Built Assuming an AI Agent Might Be the One Driving

The open-source, browser-based WGSL toolkit ships 16 built-in compute simulations and a full MCP server, so Claude Code or any other agent can write, evolve, and preview shaders the same way a human would.

ShaderVine is a browser-based WebGPU shader programming toolkit — a Monaco-powered WGSL code editor with live preview, a node-graph editor, and export targets for Unity, Unreal, Blender, Three.js, and raw HLSL. None of that alone would be remarkable in 2026’s increasingly crowded field of shader tools. What makes it worth a closer look is a design decision sitting underneath all of it: ShaderVine ships with a full MCP (Model Context Protocol) server, meaning every creative operation the editor exposes to a human is also exposed, identically, to an AI agent.

Sixteen simulations, one shared surface

Out of the box, ShaderVine includes 16 GPU compute simulations covering a genuinely broad swath of generative-art staples: Conway’s Game of Life, reaction-diffusion, physarum transport networks, fluid dynamics, falling sand, erosion, particle swarms with trails, magnetic field visualization, diffusion-limited aggregation, domain warping, and turbulence, among others. It also includes a genetic evolution system for shaders — a mechanism for mutating and selecting shader variants over generations rather than hand-tuning parameters — and visual morph transitions between states. Everything runs entirely client-side, in-browser, with no installation, and the project is MIT-licensed.

Why the MCP server is the actual pitch

The framing the project’s own writeup leads with is explicit: ShaderVine was architected to be forked and maintained with Claude Code, and built for what it calls “the agentic era.” Practically, that means an AI coding agent can open the same editor state a human has open, issue the same shader-compile and parameter-tweak operations through MCP, and hand control back and forth without translation loss. For creative coders who’ve started pairing with agents on generative work, that’s a meaningfully different experience than pasting WGSL into a chat window and copying the result back — the agent is operating the actual tool, not describing code for a human to run.

Where this fits in the current shader-tooling landscape

ShaderVine arrives alongside a wave of WebGPU-native shader tools — OpenShaders’ open-source effects directory, gpu-curtains’ DOM-synced 3D engine, and the broader maturation of Three.js’s own Shading Language (TSL) as WebGPU becomes production-ready across browsers. What distinguishes ShaderVine in that field isn’t its shader capability so much as its bet on who — or what — is going to be writing shaders alongside a human a year from now.