Shebe
github.com
- Free plan
- Windows
- macOS
- Linux
- API
- Self-hosted
What Shebe is
Shebe is a local code-search tool for repositories, using BM25 full-text search to rank matches. Queries can use phrases, Boolean operators, or literal matching. The project requires no embeddings, GPU, or cloud service, and provides 14 MCP tools for coding agents, including search, repository indexing, file reading, and reference finding. Repositories need to be cloned locally before indexing; remote repository URLs are not supported. Search is text-based rather than semantic, so a term such as “login” will not match “authenticate.” Prebuilt binaries are listed for Linux x86_64 and macOS on Apple Silicon and Intel; Linux ARM64 is planned. Windows users must build Shebe from source. Installation documentation covers MCP setup and Zed editor configuration, and points to the organization for VS Code and other editor extensions. The README reports query latency of 2 ms and indexing rates from 1,928 to 11,210 files per second in its listed benchmarks. Shebe is free, self-hosted, and listed under the Apache-2.0 license.
Who it suits
Shebe is aimed at developers and AI coding agents searching large codebases. It may suit users who want local full-text search alongside structural tools such as Serena MCP.
Good to know, and what to watch
- Ranks BM25 full-text results with phrase and Boolean queries.
- Runs locally without embeddings, GPU, or cloud service.
- Provides 14 MCP tools for coding agents.
- Prebuilt binaries are listed for Linux x86_64 and macOS.
- Free and listed under the Apache-2.0 license.
- Repositories must be cloned locally before indexing.
- Remote repository URLs are not supported.
- Search does not provide semantic similarity.
- Windows users need to build from source.
Plans
Every plan and priceFree
Free plan
The verdict
Shebe provides locally run, ranked text search and MCP tools for repository work. Its limitations are important: repositories must already be local, and related terms may not match unless the wording overlaps.
What the maker publishes
- Integrations
- The installation guide documents MCP setup and a Zed editor configuration, and points to the organization for VS Code and other editor extensions. source
- Intended users
- The project positions Shebe for developers and AI coding agents working with large codebases and says it complements structural tools such as Serena MCP. source
- License
- The GitHub organization lists the Shebe repository under the Apache-2.0 license. source
- Local processing
- The project describes Shebe as requiring no embeddings, GPU, or cloud service. source
- Maintainer location
- The GitHub organization profile lists the United States of America. source
- MCP tools
- Shebe provides 14 MCP tools for coding agents, including search, repository indexing, file reading, and reference finding. source
- Performance
- The README reports 2 ms query latency and indexing rates of 1,928 to 11,210 files per second in its listed benchmarks. source
- Purpose
- Shebe is a local code-search tool that uses BM25 full-text search for code repositories. source
- Repository requirement
- Repositories must be cloned locally before indexing, and the README says remote repository URLs are not supported. source
- Search
- Code search supports ranked results, phrases, Boolean queries, and literal matching. source
- Search limitation
- Shebe does not provide semantic similarity, so a query such as “login” does not match “authenticate.” source
- Security
- The README says Shebe runs locally and uses no cloud service; the linked security policy page could not be fetched during this research. source
- Support
- The project provides installation, configuration, troubleshooting, and MCP setup guides in its repository. source
- Supported systems
- Prebuilt binaries are listed for Linux x86_64 and macOS on Apple Silicon and Intel; Linux ARM64 is listed as planned. source
- Windows
- The installation guide says Windows users need to build Shebe from source. source
Checked against the maker's pages on 4 Oct 2026.