Ferris Search

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)

irm https://raw.githubusercontent.com/lispking/ferris-search/main/install.ps1 | iex

Or from a local clone:

.\install.ps1

> Prerequisite: Rust must be installed before running either script.

---

Quick Start

Install from source

cargo install --path .

This installs the ferris-search binary to ~/.cargo/bin/ferris-search. Make sure ~/.cargo/bin is in your PATH.

To find the installed binary path:

which ferris-search
# or
echo "$(cargo home 2>/dev/null || echo $HOME/.cargo)/bin/ferris-search"

Claude Desktop / Cursor configuration

{
  "mcpServers": {
    "ferris-search": {
      "command": "/Users//.cargo/bin/ferris-search",
      "env": {
        "DEFAULT_SEARCH_ENGINE": "bing"
      }
    }
  }
}

Replace the path with the output of which ferris-search.

Claude Code (claude mcp add)

Add for the current project only:

claude mcp add ferris-search $(which ferris-search)

Add globally for all projects (-s user):

claude mcp add -s user ferris-search $(which ferris-search)

With environment variables:

claude mcp add -s user ferris-search $(which ferris-search) \
  -e DEFAULT_SEARCH_ENGINE=bing \
  -e ALLOWED_SEARCH_ENGINES=bing,duckduckgo,brave

> -s user registers the server in your user-level config (~/.claude.json) so it is available across all projects, not just the current one.

Behind the GFW (proxy setup)

claude mcp add -s user ferris-search $(which ferris-search) \
  -e USE_PROXY=true \
  -e PROXY_URL=http://127.0.0.1:7890 \
  -e DEFAULT_SEARCH_ENGINE=bing

MCP Tools

web_search

Search the web using one or more engines simultaneously.

{
  "query": "rust async runtime",
  "engines": ["bing", "duckduckgo"],
  "limit": 10
}

| Parameter | Type | Default | Description | |-----------|------|---------|-------------| | query | string | required | Search query | | engines | string[] | server default | Engines to search (fan-out if multiple) | | limit | number | 10 | Max results per engine (1–50) |

Supported engines: bing, duckduckgo, brave, baidu, csdn, juejin, exa, firecrawl, zhihu, linuxdo, jina, tavily, github, github_code

fetch_web_content

Fetch and extract text content from any public URL.

| Parameter | Type | Default | Description | |-----------|------|---------|-------------| | url | string | required | Public HTTP/HTTPS URL | | max_chars | number | 30000 | Max characters to return |

fetch_github_readme

Fetch the README from a GitHub repository.

| Parameter | Type | Description | |-----------|------|-------------| | url | string | GitHub repository URL |

fetch_csdn_article / fetch_juejin_article / fetch_zhihu_article / fetch_linuxdo_article

Domain-specific fetchers with better content extraction than the generic fetch_web_content.

| Tool | URL Constraint | |------|----------------| | fetch_csdn_article | must contain csdn.net | | fetch_juejin_article | must contain juejin.cn and /post/ | | fetch_zhihu_article | must contain zhihu.com | | fetch_linuxdo_article | must contain linux.do and /topic/ |

Configuration

All configuration is done via environment variables.

| Env Var | Default | Description | |---------|---------|-------------| | DEFAULT_SEARCH_ENGINE | bing | Engine used when engines param is omitted | | ALLOWED_SEARCH_ENGINES | all engines | Comma-separated allow-list | | BRAVE_API_KEY | — | Required for brave engine | | EXA_API_KEY | — | Required for exa engine | | FIRECRAWL_API_KEY | — | Required for firecrawl engine | | JINA_API_KEY | — | Required for jina engine | | TAVILY_API_KEY | — | Required for tavily engine | | GITHUB_TOKEN | — | Optional for github / github_code engines (raises rate limit from 60 to 5000 req/hr) | | USE_PROXY | false | Enable HTTP/SOCKS5 proxy | | PROXY_URL | http://127.0.0.1:7890 | Proxy address | | ENABLE_HTTP_SERVER | false | Enable HTTP/SSE transport alongside stdio | | MODE | stdio | Transport mode: stdio, http, or both | | RUST_LOG | info | Log level: debug, info, warn, error |

Common configurations

Privacy-focused:

claude mcp add -s user ferris-search $(which ferris-search) \
  -e DEFAULT_SEARCH_ENGINE=duckduckgo \
  -e ALLOWED_SEARCH_ENGINES=duckduckgo,brave

Chinese developer workflow:

claude mcp add -s user ferris-search $(which ferris-search) \
  -e DEFAULT_SEARCH_ENGINE=bing \
  -e ALLOWED_SEARCH_ENGINES=bing,baidu,csdn,juejin,zhihu

With Exa AI search:

claude mcp add -s user ferris-search $(which ferris-search) \
  -e DEFAULT_SEARCH_ENGINE=exa \
  -e EXA_API_KEY=exa-xxxx \
  -e ALLOWED_SEARCH_ENGINES=exa,bing,duckduckgo

Acknowledgements

Inspired by open-webSearch — this project reimplements the same idea in Rust for performance and portability.

License

Apache-2.0

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: lispking
• Source: lispking/ferris-search
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.