Mantic.sh

Why we included this project

Most code search tools match text, which is why AI agents end up wading through false positives in a large monorepo. Mantic takes a different approach: it understands code structure, so a query like 'verify user' can find conceptually relevant code even when the exact words don't appear. Results are ranked by path structure, CamelCase detection, and optional local semantic embeddings rather than raw speed. It runs as an MCP server you can connect to Claude Desktop, Cursor, or VS Code, and it can also jump to a symbol's definition or list every reference across the repo, which is handy when an agent needs to see how a function is used before editing it. Everything runs locally, and the learned-context feature lets a team commit the search patterns that worked, so future queries build on past experience. It accepts a speed trade-off on very large codebases in exchange for much better relevance, making it a practical complement to ripgrep for teams working across big or multi-repo codebases.