Installing takes minutes; generating the per-version documentation the validator checks against is a separate step that runs an LLM over the parsed source and, by the project's own estimate, costs a couple of dollars per PennyLane version you want covered. Decide up front which versions you need — that choice is what determines whether validation can answer at all.
A Python MCP server that statically checks quantum computing code. It parses with the standard library's AST module, compiles the file, and compares every quantum library method call against documentation for a specific library version, so version-specific mistakes surface before you run anything.
- `validate_quantum_method_by_static` — takes the code and optionally a version, and reports whether the methods used exist in that release
- `request_quantum_method_reference` — returns documentation for a named method at a chosen version, so the model can correct itself
- Version-specific validation across PennyLane v0.35.0 through v0.41.1
- A three-step check: syntax, compilation, then documented method usage
- An architecture the project intends to extend to other quantum libraries
A virtual environment and uv, then `uv sync` in the cloned repo and `uv run src/server.py`. Before it is useful you must build the reference documents yourself, in two steps: parse the installed PennyLane source into JSON per version, then format that into documents with an LLM — that second step needs OPENAI_API_KEY and the project puts the cost at approximately $2.50 per version. Your client config launches it with uv from the project directory over stdio. MIT licensed.
Build from source — clone the repository and build it, then point your client at the binary