Reviews from an assistant tend toward the encouraging; this one is tuned the other way, which makes it useful for the pass where you want the nitpicks surfaced rather than smoothed over.
A code-review server with a fixed persona. You give it code, it returns a numbered critique — what is wrong, why it will bite you, and a verdict at the end — written in the voice of the senior developer who has read this pattern too many times.
- A review of the code you pass in, broken into numbered findings with a closing verdict
- Findings that name the specific construct rather than the file — the README's own example calls out a template literal used for a static string and the missing semicolon behind it
- A model you choose: set `GENAISCRIPT_DEFAULT_MODEL` to a provider and model such as `github:gpt-4o`, or leave it on the default
Node.js 20+ on your machine; the server runs as `npx -y grumpydev-mcp` over stdio, and one-click install links plus `code --add-mcp` lines are in the README. The review is generated by a model, so the provider matters: by default it uses GitHub Models with GPT-4o, and other providers configured through GenAIScript may need their own API keys set as environment variables — point `envFile` at a `.env` if you keep them in one. In VS Code, MCP support requires Agent mode to be enabled.
One command — npx -y grumpydev-mcp