Rust로 작성된 다중 엔진 웹 검색을 위한 blazing-fast MCP 서버.

왜 ferris-search?

Claude Code’s built-in web search works great in ideal network conditions — but in practice, many developers run into environment where it’s unreliable or unavailable: 기업 네트워크, 제한적 영역, 공기 승인 설정, 또는 단순히 스포트티 연결.

workaround를 찾고 있지만, 여러 엔진을 통해 쿼리를 검색하는 Node.js MCP 서버에서 open-webSearch를 통해 왔습니다. 그것은 문제를 잘 해결했습니다. 그러나 Rust에 대한 일이 있습니다. — 그리고 Node.js 런타임을 거의 HTTP 요청이 필요한 것보다 무거운 것을 느꼈습니다.

그래서 나는 Rust에서 같은 아이디어를 rewrote :

• **No Node.js 런타임 ** - 단일 자체 유지 바이너리, ~8 MB -Lower latency — Rust async I/O, 엔진의 동시 팬 아웃
• ** 소형 ** — negligible 메모리 사용
• ** 프록시 지원** - HTTP/SOCKS5 프록시를 통해 env var, 필요한 네트워크

Claude Code의 검색이 환경에 작동하지 않는 경우, 이것은 당신을 위해 것입니다.

기업 & 내부 사용

ferris-search는 ** 내부 검색 ** 시나리오에 대한 좋은 기반입니다.

ferris-search 🦀

A blazing-fast MCP server for multi-engine web search, written in Rust.

Why ferris-search?

Claude Code’s built-in web search works great in ideal network conditions — but in practice, many developers run into environments where it’s unreliable or unavailable: corporate networks, restricted regions, air-gapped setups, or simply spotty connectivity.

While looking for a workaround, I came across open-webSearch, a Node.js MCP server that routes search queries through multiple engines. It solved the problem well. But I have a thing for Rust — and spinning up a Node.js runtime just to proxy a few HTTP requests felt heavier than it needed to be.

So I rewrote the same idea in Rust:

• No Node.js runtime — single self-contained binary, ~8 MB
• Lower latency — Rust async I/O, concurrent fan-out across engines
• Smaller footprint — negligible memory usage
• Proxy support — HTTP/SOCKS5 proxy via env var, for networks that need it

If Claude Code’s search isn’t working in your environment, this is for you.

Enterprise & Internal Use

ferris-search is also a good foundation for enterprise internal search scenarios. Since it’s a standard MCP server written in Rust, you can fork it and add custom search engines that connect to your internal knowledge bases — Confluence, Notion, internal wikis, code repositories, or proprietary document stores.

Some ideas:

• Add an engine that searches your internal Elasticsearch or OpenSearch cluster
• Integrate with your company’s Confluence or GitLab search API
• Connect to a private RAG (Retrieval-Augmented Generation) service
• Route queries to different backends based on query language or topic

With Claude Code as the AI layer and ferris-search as the search backbone, your team gets a local AI coding assistant that can actually find and reference internal documentation — without sending anything to external search engines.

Features

• Multi-engine fan-out — search across multiple engines simultaneously with a single call
• 14 search engines — Bing, DuckDuckGo, Brave, Baidu, CSDN, Juejin, Exa, Firecrawl, Zhihu, LinuxDo, Jina, Tavily, GitHub (repo search), GitHub Code (code search)
• 7 MCP tools — web_search + 6 article/content fetchers
• No API keys required for most engines (Brave, Exa, Firecrawl, Jina, and Tavily require API keys)
• Single binary — ~8 MB, no runtime dependencies
• Proxy support — HTTP/SOCKS5 proxy via env var

Quick Install

Linux / macOS (one-liner)

curl -fsSL https://raw.githubusercontent.com/lispking/ferris-search/main/install.sh | bash

Or from a local clone:

bash install.sh

The script will:

1. Build and install the binary via cargo install
2. Register the MCP server with Claude Code (claude mcp add -s user) if the CLI is found
3. Install Claude Code skills for ferris-search

Windows (PowerShell)