Summary

mgrep provides a CLI-native semantic search for code, images, and PDFs, overcoming grep's limitations. It targets developers and AI agents, enabling immediate, natural-language querying that reduces agent token usage and enhances search relevance.

How It Works

Utilizing Mixedbread Search, mgrep combines semantic retrieval models with optimized inference. mgrep watch continuously indexes repositories in the background, respecting .gitignore, and stores data in a cloud-backed corpus. Users query this corpus via natural language, facilitating intent-based discovery beyond traditional pattern matching and allowing models to focus on reasoning.

Quick Start & Requirements

Install via npm install -g @mixedbread/mgrep (requires Node.js/npm/pnpm/bun). Authenticate via mgrep login (browser) or MXBAI_API_KEY env var. Index projects with mgrep watch in the repository directory. Search using mgrep "natural language query" [path]. Claude Code integration is available via mgrep install-claude-code.

Highlighted Details

• Semantic, multilingual search; multimodal (audio, video) support is upcoming.
• Achieved ~2x agent token reduction vs. grep-based workflows in benchmarks, with comparable or better quality.
• Features first-class coding agent integrations, starting with Claude Code.
• Cloud-backed stores allow shared querying without re-uploading data.

Maintenance & Community

No specific details on maintainers, community channels, or roadmaps were provided in the README.

Licensing & Compatibility

Licensed under Apache-2.0, offering permissive terms for commercial use and closed-source integration.

Limitations & Caveats

Audio/video support is pending. Login issues may require mgrep logout. mgrep watch can be noisy; consider --store for workspace separation or pausing the watcher. Automated tests are not fully integrated; pnpm typecheck is recommended pre-publish. mgrep complements, rather than replaces, grep.