Plenty of assistants will write you a pytest file; the difference here is that `run_tests` executes it in an isolated environment with a timeout and hands back coverage, so a generated suite gets checked before you commit it. Everything is passed as code strings rather than file paths, which keeps it self-contained but means you are feeding it snippets, not pointing it at a repository.
A Python-focused server that closes the loop between reviewing code and proving it works. It runs the standard static analysis stack over a snippet, generates a pytest suite from the code's own AST — happy path, edge cases, exceptions and type checks — then executes that suite in an isolated environment and reports coverage.
- `code_review` runs pylint, flake8, mypy, bandit, radon and AST inspections and returns issues, security findings and complexity metrics
- `generate_tests` builds a pytest suite from a function or class, covering happy path, edge cases, exceptions and type validation
- `run_tests` executes tests with a timeout and returns pass/fail, coverage percentage and line coverage
- `format_code` reformats to PEP8 with black or autopep8 at a line length you choose
Python 3.8+ and a checkout of the repository — install it with `pip install -e .` (package `neurodev-mcp`, version 0.2.0), preferably into a virtualenv, then run `python -m neurodev_mcp.server` or the `neurodev-mcp` command. Stdio is the default transport. The analyzers it shells out to are the usual Python toolchain, so they need to be installed alongside it.
One command — pip install -e .