Awesome MCP › Software Development

xerj-org/xerj

⭐ 18616 Rust added to this list on 2026-11-02 repository created 2026-06-30

XERJ is a local-first search and indexing engine for AI coding agents, distributed as a single static Rust binary with no JVM or external services. It automatically indexes a project's source code, documentation, specs, logs, JSONL/CSV data and PDFs, then serves BM25 full-text search, vector (kNN) search and hybrid ranking behind an Elasticsearch-compatible API. A native MCP server (`xerj init` wires it into Claude Code or Cursor as both an MCP tool and a skill) lets an agent call `xerj_batch_search` and related tools directly, pulling cited, file:line-accurate passages instead of reading files one at a time. The headline use case is "reference coding": pointed at a task, XERJ clones and indexes the open-source repositories closest to what is being built and lets the agent reuse how those projects solved the same problem before writing its own code. The maintainers report a controlled study cutting a coding agent's output tokens by 2.7x at an unchanged solve rate, and users report roughly 5x savings in everyday use. Beyond coding, XERJ also exposes a `/_memory` namespace agents can use as persistent memory across sessions, and a `xerj def` command for go-to-definition-style lookups by symbol name. The default embedder is lexical and fully offline; a neural embedder is opt-in and downloads about 90 MB of weights. By default the server binds to localhost with API-key auth; an `--insecure` flag disables TLS and auth entirely and is meant only for one trusted user on one machine, since any caller that can reach the port then has full read and delete access to every index. XERJ is Apache-2.0 licensed and asks each user, human or agent, for one short field report on how it was used in return for the free binary.

https://github.com/xerj-org/xerj

searchindexingcode-searchagent-memoryrustbm25hybrid-searchreference-coding