Bring real-time internet access to your Claude, ChatGPT, or local LLM — for free.

Question

I Built an Open-Source Web Search Plugin for AI Agents — No API Key Needed

Bring real-time internet access to your Claude, ChatGPT, or local LLM — for free.

---

As someone who builds AI assistants and tools, I’ve always hit the same wall:
Language models are smart — but blind to the current internet.

Sure, Claude has a browsing plugin, and GPT-4 can use tools… but:

• They're paid features (Bing/GCS API isn’t cheap),
• You often can’t customize the results or search logic,
• And some platforms don’t allow plugin access at all (especially for local or self-hosted models).

So I built an open-source alternative — Open-WebSearch MCP — a lightweight, multi-engine search tool that speaks the MCP protocol (used by Claude, LangChain, etc).

---

What It Does

✅ Multi-engine search with no API keys
Currently supports:
Bing, Baidu, CSDN, DuckDuckGo, Exa, Brave
(And more like Reddit, GitHub, Google are coming.)

✅ Structured output for RAG
Each result includes title, url, snippet, engine, and source. Easy to parse for chat agents and document systems.

✅ Article content extraction
Want to fetch full blog posts from CSDN or Linux.do for summarization or RAG pipelines? Built-in support.

✅ Supports Claude, Cherry Studio, LangChain, and more
Implements both streamableHttp and SSE MCP transports.

✅ Free & self-hosted
No rate-limited APIs. No account needed. Fully local control over what and how you search.

---

How to Use

Install locally:

git clone https://github.com/Aas-ee/open-webSearch
cd open-webSearch
npm install
npm run build
node build/index.js

Or use Docker:

docker run -d --name web-search \
  -p 3000:3000 \
  -e ENABLE_CORS=true \
  ghcr.io/aas-ee/open-web-search:latest

---

Example: Claude Plugin Config (Cherry Studio)

{
  "mcpServers": {
    "web-search": {
      "name": "Web Search MCP",
      "type": "streamableHttp",
      "baseUrl": "http://localhost:3000/mcp"
    },
    "web-search-sse": {
      "name": "Web Search SSE",
      "type": "sse",
      "url": "http://localhost:3000/sse"
    }
  }
}

---

Tools It Provides

search → perform multi-engine search

use_mcp_tool({
  server_name: "web-search",
  tool_name: "search",
  arguments: {
    query: "how to build a Claude plugin",
    engines: ["bing", "csdn"],
    limit: 5
  }
})

fetchCsdnArticle → fetch full blog text
fetchLinuxDoArticle → (experimental) get Linux.do article content

---

⚙️ Dev Features

• Supports HTTP proxies (USE_PROXY, PROXY_URL env vars)
• Configurable default engine
• Auto builds Docker images via GitHub Actions
• Easy to fork + customize for your own toolchains

---

⭐ Try It, Fork It, Improve It

GitHub: https://github.com/Aas-ee/open-webSearch
Docker: ghcr.io/aas-ee/open-web-search:latest

If you’re building an LLM agent, this gives your assistant eyes on the internet. No API keys. No paywalls.

Let me know what engines or features you’d like next!

---

Happy hacking — and let your LLMs browse freely!