XERJ

•
19k
•

Index codebases, documents, and structured data into a local-first search engine and MCP server designed for AI agent memory, RAG, and fast code retrieval.

XERJ is a high-performance, local-first AI search engine and MCP server distributed as a single Rust binary. Designed to eliminate expensive grep operations and token-heavy context stuffing for AI coding agents, it automatically indexes source code, documents, logs, JSONL/CSV files, and PDFs for BM25 full-text, vector, and hybrid search. Featuring native Model Context Protocol (MCP) support and an Elasticsearch-compatible API, XERJ empowers coding tools like Claude Code and Cursor to execute precise reference coding, retain structured agent memory, and run fast, verified RAG queries entirely on local CPU.

主な機能

01Elasticsearch-compatible REST API packaged within a lightweight, JVM-free single binary
0219,181 GitHub stars
03Zero-configuration auto-indexing for codebases, PDFs, logs, JSONL/CSV, and SQLite databases
04Native Model Context Protocol (MCP) server integration compatible with Claude Code, Cursor, and custom bots
05Multi-modal search engine supporting BM25 full-text, kNN vector, and hybrid search algorithms
06XERJ Query-1 execution engine for offline, deterministic CPU-based question answering and fact verification

ユースケース

01Enabling AI coding agents to perform reference coding by pulling exact code passages from indexed repos without burning context tokens
02Providing local-first persistent memory, structured facts, and RAG capabilities for autonomous agents and bots
03Executing fast semantic and syntactic code searches and definition lookups across massive multi-language repositories