Hey everyone 👋 I'm Ameya, one of the makers behind Mustel, along with Raunak Nayak. We built this because we kept noticing the same thing: our AI coding agents (Cursor, Claude Code, Windsurf) were burning tons of context just re-reading files and parsing raw linter output on every save, and still occasionally missing bugs a plain linter would've caught instantly, or hallucinating a fix for something that wasn't actually broken. So Mustel is our attempt at fixing that layer specifically. It's a local, non-AI static analysis tool. It runs Ruff, Bandit, and pip-audit under the hood, dedupes the overlapping findings, and compresses everything into a single short agent_prompt your AI agent can act on directly instead of parsing a wall of logs. It also ships an MCP server, so it auto-registers with Cursor, Windsurf, Claude Code, and Claude Desktop in one command. Some numbers from our benchmarks: -34.4% context tokens, -64.9% output tokens, and it caught real vulnerabilities across our test projects (75% recall on code it wasn't tuned on). It's fully open source (MIT), on PyPI (`pip install mustel`), and still early: 19 stars, actively maintained, lots we want to add next (more language coverage, smarter rule tuning). Would genuinely love feedback, bug reports, or just "this rule is dumb, fix it" type comments. Happy to answer anything about how it works under the hood too. Thanks for checking it out 🙏

Mustel

Non-AI static analysis, built for AI coding agents

Non-AI static analysis, built for AI coding agents