semble

Semble is a Python MCP server that provides semantic code search for AI coding agents, indexing a full repository in roughly 250 milliseconds and returning query results in about 1.5 milliseconds, entirely on CPU with no API keys or GPU required. It integrates with Claude Code, Cursor, Codex, OpenCode, and any other MCP-compatible agent through three deployment modes: an MCP server tool, CLI instructions injected into AGENTS.md or CLAUDE.md, and a dedicated sub-agent named semble-search. An interactive installer detects which coding agents are present and configures the chosen integrations automatically. Rather than passing full file contents to the agent, Semble returns only the relevant code chunks, which the project benchmarks at roughly 98 percent fewer tokens than a grep-plus-read approach while achieving an NDCG@10 of 0.854, comparable to code-specialized transformer models. The CLI supports natural-language queries against local paths or remote git URLs, a find-related command for locating code similar to a known file and line, and a savings command that displays cumulative token savings across all searches.

Lo que la gente pregunta sobre semble