SoSearch MCP server
Commit Oct 16, 2026Aggregates DuckDuckGo, Brave, and Yahoo search results into a standardized format for AI agents via HTTP API or MCP.
Also known as NetLops/SoSearch
- Stars
- 135
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Oct 16, 2026
- License
- —
What is SoSearch MCP?
SoSearch aggregates DuckDuckGo, Brave, and Yahoo search results into a standardized format for AI agents via HTTP API or MCP.
Summary generated from the project's README
About SoSearch MCP server
SoSearch provides a way for developers and AI agents to access web search results without paying for expensive third-party API keys. It works by multiplexing requests to DuckDuckGo, Yahoo, and Brave Search, then normalizing the HTML results into a single standardized JSON format. This tool is built with Rust and uses asynchronous I/O to handle multiple requests concurrently.
Users can deploy this as a local HTTP API or run it as an MCP server using the --mcp flag. When used in MCP mode, it exposes a web_search tool that integrates with clients like Claude Desktop, Gemini CLI, or Cursor. The server uses TLS impersonation to simulate a Chrome browser footprint, which helps minimize blocking during the scraping process. It is designed for local execution via pre-built binaries or by compiling from source with Cargo.
What you can do with it
- 1Perform web searches via AIAn AI agent can use the web_search tool to find real-time information from multiple search engines.
- 2Avoid expensive search APIsDevelopers can emulate services like SerpAPI or Tavily without needing official API keys.
Written from the project's README. Every use case is checked against a quote from it.
Automated checks
Ran Nov 14, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachablenetlops/sosearch is public and reachablegithub.com/netlops/sosearch ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)Not checked
- Open source license (OSI)No licenseFinding
- Commit or release in the last yearLast commit or release 2026-10-16github.com/netlops/sosearch ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNot checked
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model0 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNot checked
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubSoSearch API
A lightning-fast pseudo Web Search Engine API written in Rust — also works as an MCP server for AI agents. This project emulates popular APIs like SerpAPI or Tavily without needing official and expensive API keys, by multiplexing requests to popular engines directly and scraping the results concurrently.
Philosophy
- Performance: Powered by
tokiofor async concurrent I/O. - Bot Bypass: Leverages
rquestwith TLS impersonation (e.g., simulating a Chrome 124 browser footprint at the TLS/HTTP2 layer) to minimize blocking vs standard HTTP clients (the Rust equivalent ofcurl_cffi). - Standardized: Normalizes
DuckDuckGo,Yahoo, andBraveHTML results into a standardizedSearchResultJSON array.
Core Stack
🔍 Supported Search Engines
- DuckDuckGo (Primary standard search)
- Yahoo (Powered by Bing)
- Brave Search (Independent index)
📁 Project Structure
src/
├── main.rs # Entry point: HTTP server or MCP mode (--mcp)
├── search.rs # Shared concurrent search logic
├── mcp.rs # MCP stdio server (JSON-RPC 2.0)
├── models.rs # SearchResultItem, SearchResponse structs
└── engines/
├── mod.rs # SearchEngine enum + trait dispatch
├── duckduckgo.rs # DuckDuckGo scraper
├── yahoo.rs # Yahoo scraper (Bing-powered)
└── brave.rs # Brave Search scraper
examples/
├── fetch_html.rs # Download raw HTML for offline debugging
└── test_parser.rs # Offline CSS selector iteration
.gemini/ # Gemini CLI agent config
├── GEMINI.md # Project-level system prompt
├── settings.json # MCP server configuration
└── skills/ # Project-level agent skills
├── sosearch-engine-dev/ # Scraper development workflow
└── sosearch-api-ops/ # API operations & deployment
.agents/ # Generic agent config (compatible with multiple AI tools)
└── skills/ # Same skills, alternative discovery path
├── sosearch-engine-dev/
└── sosearch-api-ops/
🤖 Agent Skills & MCP Support
This project includes built-in AI agent support for both Gemini CLI and other tools that follow the .agents/ convention.
Available Skills
| Skill | Description |
|---|---|
sosearch-engine-dev |
Full workflow for adding/debugging search engine scrapers: fetch HTML → test selectors offline → decode URLs → integrate |
sosearch-api-ops |
Operations guide: build, run, test, deploy (local + Docker), troubleshoot |
MCP Servers
Configured in .gemini/settings.json:
| Server | Package | Purpose |
|---|---|---|
filesystem |
@modelcontextprotocol/server-filesystem |
Scoped file access to project directory |
Usage with Gemini CLI
cd /path/to/SoSearch
gemini
# Skills are auto-discovered. Ask: "How do I add a new search engine?"
🔌 MCP Server Mode
Run SoSearch as an MCP server for AI agents (Claude, Gemini, Cursor, etc.):
./SoSearch --mcp
Configuration
Claude Desktop (claude_desktop_config.json):
{
"mcpServers": {
"sosearch": {
"command": "/path/to/SoSearch",
"args": ["--mcp"]
}
}
}
Gemini CLI (.gemini/settings.json):
{
"mcpServers": {
"sosearch": {
"command": "/path/to/SoSearch",
"args": ["--mcp"]
}
}
}
This exposes a web_search tool that AI agents can call to search the web.
Windows Configuration
Claude Desktop (%APPDATA%\Claude\claude_desktop_config.json):
{
"mcpServers": {
"sosearch": {
"command": "C:\\path\\to\\SoSearch.exe",
"args": ["--mcp"]
}
}
}
🚀 Quick Start
Linux / macOS
# Download pre-built binary or build from source
cargo run --release
curl "http://localhost:10080/search?q=hello+world"
Windows
使用预编译二进制:
从 GitHub Releases 下载 SoSearch-windows-amd64.zip,解压后:
# 启动 HTTP 服务
.\SoSearch.exe
# 另开一个终端测试
Invoke-RestMethod "http://localhost:10080/search?q=hello+world" | ConvertTo-Json
# 或使用 curl
curl.exe "http://localhost:10080/search?q=hello+world"
从源码编译(需要安装 Rust、CMake、NASM、LLVM/Clang):
# 安装依赖 (使用 Chocolatey)
choco install cmake nasm llvm -y
# 编译运行
cargo run --release
MCP 模式:
.\SoSearch.exe --mcp
Refer to QUICK_START.md for full instructions.
📄 License
CC BY-NC 4.0 — 非商业用途可自由使用、修改和分发。
🏠 Community
首发于 LINUX DO 社区,欢迎 Star ⭐ 和 PR!
SoSearch MCP questions
Do I need an API key to use SoSearch?
No, the service does not require any official or expensive API keys.
How can I run SoSearch as an MCP server?
You can run it by executing the binary with the --mcp flag.
What technologies is SoSearch built with?
It is written in Rust and utilizes the Axum, Tokio, rquest, and scraper libraries.