About GrowBlocks Dev Kit
GrowBlocks Dev Kit helps AI assistants build, test, and submit three.js browser games. It combines headless play-testing with screenshots, reusable community blocks, and a review-based publishing flow.
Key features
- 01Read Dev Kit API documentation
- 02Create game projects and run static checks
- 03Play headlessly and return screenshots
- 04Find and add published blocks
- 05Upload unlisted builds and submit for review
Use cases
- 01Start a three.js browser game from a starter project
- 02Iterate on game changes using play-test screenshots
- 03Reuse blocks published by other developers
- 04Share a test build and submit a game for review
MCP support
The MCP server provides Claude, Cursor, Windsurf, and other clients with tools for game documentation, checks, blocks, and publishing, along with prompts and resources.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
Tools
16Saved tool list · updated 2w ago
| Parameter | Type | Required | Description |
|---|---|---|---|
| dir | string | No | The game folder (default: the folder the assistant runs in). |
| topic | string | No |
Details
From the project READMEGrowBlocks Dev Kit for Claude, Cursor and other AI assistants
Make a browser game with an AI assistant and ship it on GrowBlocks: three.js, Rapier physics, characters, multiplayer rooms and hosting, with no setup.
The assistant reads the API, writes game.js, plays the game headless and looks at a screenshot after every change, reuses parts other developers published (blocks), and pushes an unlisted test build you can share. Nothing goes public until you submit it and GrowBlocks reviews it.
Install
Claude Code (plugin: tools + skill)
/plugin marketplace add Gioluan/growblocks-devkit
/plugin install growblocks@growblocks
Claude Code (just the MCP server)
claude mcp add growblocks -- npx -y growblocks mcp
Claude Desktop: download growblocks.mcpb from the latest release and open it.
Cursor / Windsurf / any MCP client
{ "mcpServers": { "growblocks": { "command": "npx", "args": ["-y", "growblocks", "mcp"] } } }
Terminal: npx growblocks new my-game, then npx growblocks help.
Then ask: "Make me a GrowBlocks game where frogs race across lily pads", or use the make_game prompt.
Tools
| Tool | What it does |
|---|---|
growblocks_docs | The Dev Kit API (or one section of it) |
growblocks_check | Static checks; with play it plays the game headless and returns a screenshot |
growblocks_new_game | A new game folder from the starter kit |
growblocks_blocks, growblocks_block_info, growblocks_add_block | Find and add published blocks (their authors earn when your game ships them) |
growblocks_login, growblocks_whoami | Sign in with a code you confirm in your browser |
growblocks_create_game, growblocks_link_game, growblocks_update_game | Make the game on GrowBlocks and set its name, logo and details |
growblocks_push | Upload an unlisted test build and get its link |
growblocks_status, growblocks_games | Builds, links and the review note |
growblocks_submit, growblocks_publish_block | Only when you ask: submit for review, publish a block |
Every tool takes dir, the game folder. Prompts: make_game, ship_game. Resources: llms.txt, gb.d.ts.
Money (collections, the coin, earnings) never goes through the assistant: it stays in your browser at growblocks.xyz. Developers keep 70% of the fees their game generates.
Layout
.claude-plugin/marketplace.json: this repo as a Claude Code plugin marketplaceplugins/growblocks/: the plugin (.mcp.jsonrunsnpx [email protected] mcp, pinned so every install runs the reviewed version, plus thegrowblocks-gameskill)desktop/: the Claude Desktop extension; build it withnpx @anthropic-ai/mcpb pack desktop growblocks.mcpbserver.json: the entry for the official MCP RegistryTERMS.md: the Dev Kit terms
The server is one dependency-free file, growblocks.mjs, published on npm as growblocks.
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.