AI coding agents write code without understanding the codebase they work in. They miss how files connect, which code is risky to change, and why the architecture looks the way it does. The result is code that breaks things or ignores patterns the team already follows. Repowise fixes that. It maps your codebase and serves that map to AI coding agents over MCP. The agent queries Repowise before it writes, so it works with real context instead of guessing. Repowise has five layers: Graph: parses your code into a dependency graph and finds the files that matter and how changes ripple through them. Git: reads your commit history to surface hotspots, code ownership, co-change patterns, and bus factor. Docs: generates a searchable wiki of how your codebase works. Decisions: captures the reasoning behind the architecture, so changes don't quietly break it. Code Health: scores every file from 1 to 10 across 25 deterministic biomarkers, including cyclomatic complexity, duplication, and untested hotspots. It runs without an LLM, so the score is reproducible and tracks over time. You reach all of it through 9 MCP tools. Repowise works with Claude Code, Cursor, and any MCP client. Repowise is open source under AGPL-3.0. Commercial licensing is available. Star us on GitHub: https://github.com/repowise-dev/repowise Docs: docs.repowise.dev