Probe
Probe is a Rust code search and context engine that uses AST-aware parsing and ranking, with CLI, Node.js SDK and MCP modes for AI coding assistants.
- GitHub stars
- 721
- Last commit
- yesterday
- Latest release
- v0.6.0-rc341
- Licence
- Apache-2.0

Probe is a code and markdown context engine with a built-in agent, made to work on large codebases. It aims to give AI coding assistants and developers complete, relevant context by treating code as structured code rather than text, combining ripgrep-style speed with tree-sitter AST parsing.
Unlike plain text search tools, it returns whole functions or classes, and unlike embedding-based tools it needs no indexing step or embedding model. Searches use Elasticsearch-style boolean queries ranked with BM25. Probe can be used in several modes: as a Probe Agent exposed over MCP, as raw MCP tools, as a CLI agent, through direct CLI commands or from a Node.js SDK. An LLM scripting feature and a list of supported languages are covered in the documentation.
The project is written in Rust and licensed under Apache-2.0. The authors present it as useful for spec-driven development, code reviews and onboarding, where understanding code comes before writing it. It runs locally on your own machine, with LLM access configured through environment variables.
Key features
- AST-aware code search with tree-sitter
- Returns complete functions and classes
- BM25-ranked boolean queries
- No indexing or embedding setup
- MCP tools and agent mode
- CLI and Node.js SDK
Pricing: The open source engine is free to run on one node. Business costs $40 per seat per month billed annually, counting only users active in the last 30 days; Enterprise is quoted on request.