Shebe

Shebe is a self-hosted search utility for code repositories, built for developers and AI coding agents that keep source locally. It is a strong choice for fast, literal lookups with agent integration, but a poor fit for semantic discovery or workflows built around remote repositories.

Overview

Shebe indexes repositories that have been cloned onto the local machine and applies BM25 full-text ranking. Processing stays local: the tool requires no embeddings, GPU, or cloud service. That combination suits organizations that want source and search activity under their own control, while putting the indexing burden on each user or host.

Queries can use phrases, Boolean operators, or literal matching, so developers can move from broad term searches to tightly constrained lookups. Published benchmarks report 2 ms query latency and indexing rates from 1,928 to 11,210 files per second. Those figures make Shebe attractive for repeated searches in large trees, but its vocabulary is literal. A query for “login” will not automatically find code that calls the operation “authenticate.”

Key features

• Ranked code search: BM25 orders likely matches, while phrase, Boolean, and literal modes help locate known names and wording. There is no semantic similarity and no regex search, limiting concept-driven exploration and pattern matching.
• MCP toolkit: Fourteen MCP tools cover searching, repository indexing, file reading, and reference finding. Coding agents therefore get a defined interface to a local codebase rather than only a standalone search command.
• Navigation and editor links: Code navigation is supported, and the installation material documents MCP setup plus a Zed configuration. The project also points to its organization for VS Code and other editor extensions, making it more practical in integrated development environments.
• Local-first operation: No cloud service, embeddings, or GPU are needed. This reduces infrastructure dependencies and keeps repository processing self-hosted, though users must maintain their own clones and deployment.

Pricing

Shebe has one Free plan at free. It includes local code search and is Apache-2.0 licensed. There is no free trial because the product is already free, and no paid tier, seat charge, usage quota, or renewal term is published. The trade-off is operational rather than commercial: you supply the machine, local repository clones, and installation work.

Platforms

Shebe is offered as an API and supports Linux, macOS, Windows, and self-hosted deployment. Prebuilt binaries are listed for Linux x86_64 and macOS on both Apple Silicon and Intel. Linux ARM64 is planned rather than listed as a current build. Windows users must compile from source, so Windows is a less convenient choice than the platforms with downloadable binaries.

Who it's for

Shebe fits developers working across sizable local codebases and teams building AI coding-agent workflows around MCP. It is particularly well suited to users who value local processing, need quick exact-term searches, and can clone repositories before indexing. It also complements structural tools such as Serena MCP instead of trying to replace them.

Look elsewhere if repositories cannot be stored locally, if a hosted index is required, or if users routinely search by intent rather than by the words used in source. Windows teams that do not want a source build should also weigh the installation cost.

Pros and cons

Pros

• Fast local lookups: Reported 2 ms query latency targets responsive work in large repositories.
• Useful query controls: Phrase, Boolean, and literal modes help refine searches for exact code terms.
• Agent-ready interface: Fourteen MCP tools include search, indexing, file access, and reference finding.
• Low infrastructure demands: No embeddings, GPU, or cloud service is required.
• Permissive licensing: The repository is listed under Apache-2.0.

Cons

• No semantic matching: Different terminology, such as “login” versus “authenticate,” will not be connected automatically.
• Local clones are mandatory: Remote repository URLs cannot be indexed directly.
• No regex search: Pattern-oriented searches need another tool or a different workflow.
• Windows source build: Windows installation requires compiling Shebe yourself.
• Uneven binary coverage: Linux ARM64 is planned, not a current prebuilt target.

Alternatives

Code Search Tools is the broader directory for comparing code-search products. Choose Code Search instead when you need a freemium Community plan with unlimited users, find-and-replace across repositories, a CLI and MCP server, and SCIP code intelligence. ast-grep is a better direction when its API, extension, web, and desktop platform coverage suits your workflow.

GNU GLOBAL is another free option for traditional source-code tagging, while OpenGrok offers a free, open-source web-oriented cross-reference system. Pick sqry when a locally running, MIT-licensed tool is the priority. LXR Cross Referencer suits users seeking a self-installed web cross-reference application. Chromium is a general open-source browser rather than a direct code-search replacement. Chatterbox TTS addresses speech generation, not repository search.

Verdict

Choose Shebe if you want a free, Apache-2.0 local search layer for cloned repositories and MCP-enabled coding agents, especially when fast exact-term retrieval matters. Look elsewhere for semantic similarity, regex queries, remote indexing, or a no-build Windows installation. Its focused scope is a strength for disciplined local workflows, but it is not a complete answer to exploratory code understanding.