Teach any coding agent the real s&box API: 17 source-verified reference files plus an 18-tool editor MCP toolset. Stops agents writing Unity code into Source 2 projects.
Nothing in s&box tells you when you are wrong.
In short. Two things for building s&box games, either usable on its own.
The skill is 17 reference files that teach a coding agent the real s&box API, so it stops writing Unity code into your Source 2 project. Plain markdown, no runtime, no dependencies, so it works with any agent that can read a file. Every API in it is traceable to engine source at a named version.
The toolset adds 18 tools to the editor's MCP server. It answers the questions the editor otherwise leaves silent: does this type actually exist, has the compiler noticed my edit, why did nothing happen when I changed that config. Install it as a package reference, fobiat.sboxmcpserver, or drop the single C file into your Editor/ folder.
On the two words. The asset.party listing is called s&box MCP Server & AI Skill. Strictly the second half is a toolset: s&box ships the MCP server inside the editor, on 127.0.0.1:7269, and this registers into it rather than starting anything. Both words appear below and they mean different things: the server is the editor's, the toolset is what this adds to it.
You write a component. It compiles. You press play. Nothing moves.
There is no error. There is no warning. The console is clean. Somewhere between the code you wrote and the thing you expected, a call that looked completely correct did nothing at all, and the engine did not think that was worth mentioning.
From the project README.
Add the radar badge to your README — it shows your project was picked up by MCP Radar and links to this page:
[](https://mcp.liqiwa.com/s/fobiat--sbox-skill.html)
Annotated screenshots for AI agents: Skitch-style arrows, numbered steps, boxes, labels, solid redaction
replenhq/replenSmarter AI Development Workflows
MohammadHijjawi97/since-cutoffFind which APIs of your pinned Python dependencies changed after your coding model's training cutoff, and give the agent short AGENTS.md / CLAUDE.md notes from a static API diff. No model calls, no AP
shitianfang/jev-useClaude Code / Codex / pi plugin that hands agent steps needing no text output to Jev (TypeSafe's judgment model) — measured p50 ~230 ms and ~$0.02 per 1,000 judgments, with typed escalation back to th
00200200/usagetrimKeep the signal. Cut the noise. Local CLI + MCP — fold verbose tool output for Claude Code, Codex, Cursor & Desktop.
00200200/tokencutKeep the signal. Cut the noise. Local CLI + MCP — fold verbose tool output for Claude Code, Codex, Cursor & Desktop.
The top new MCP servers of the week, every Monday. No spam, unsubscribe anytime.