semble

What it does

Semble is a code search engine designed for agents. It accepts natural-language queries about a codebase and returns only the relevant code snippets, using roughly 98% fewer tokens than traditional grep-and-read approaches. The tool indexes local or remote repositories on CPU with no external services, API keys, or GPU required. Search latency averages 1.5 milliseconds per query after an initial index build, which takes about 250 milliseconds for a typical codebase. It operates as an MCP server compatible with Claude Code, Cursor, Codex, and similar agents, as well as a standalone CLI tool.

Who it's for

Agents and developers using Claude Code, Cursor, Codex, or other MCP-compatible tools who need fast code discovery. Particularly useful when working with large or unfamiliar codebases where token budgets are tight or latency matters. The token-efficiency gain translates directly to lower API costs and faster agent responses in token-constrained environments.

Common use cases

• Query a local codebase in natural language and get exact snippets without manual file hunting.
• Integrate Semble as an MCP tool in Claude Code or Cursor to enable agents to search repositories during code analysis.
• Search remote repositories by git URL, cloning them on demand for one-shot queries.
• Find code similar to a known file and line number via the find-related subcommand.
• Index and search documentation, configuration, or other non-code content alongside source code.

Setup pitfalls

• Filesystem access required: Semble reads and indexes all files matching your codebase patterns; index data is cached locally. Verify your sandboxing or permissions align with this.
• Index caching: Indexes are built automatically on first run and cached; changes to tracked files invalidate the cache. This is usually transparent but can be unintuitive if index freshness is critical.
• .sembleignore and .gitignore rules: Files excluded from git are also excluded from indexing by default. Override with .sembleignore if you need to index files git ignores.
• Network calls: Remote git URLs require network access; ensure your environment allows git clones if you plan to search repositories by URL.