DuckDuckGo Web Search MCP Server

@kouui

About DuckDuckGo Web Search MCP Server

MCP server of web search/fetch functionality using duckduckgo and jina api. no api key required.

Config

Add this server to your MCP-compatible client using the configuration below.

{
  "mcpServers": {
    "web-search-duckduckgo": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/kouui/web-search-duckduckgo.git@main",
        "main.py"
      ]
    }
  }
}

Tools

No tools detected

Fetch the live tool list by running this server in a temporary sandbox using the button above.

Overview

What is DuckDuckGo Web Search MCP Server?

An MCP (Model Context Protocol) server that enables web search through DuckDuckGo and optionally fetches and summarizes the content of found URLs. It is designed for use with any MCP‑compatible client, such as Claude Desktop.

How to use DuckDuckGo Web Search MCP Server?

Install the server via uvx and configure it in your MCP client’s configuration file (e.g., claude_desktop_config.json). Two configuration methods are provided: one using git+https and another using a local directory. Once configured, you can invoke the search_and_fetch tool (with parameters query and limit) or the fetch tool (with parameter url).

Key features of DuckDuckGo Web Search MCP Server

Web search using DuckDuckGo
Extracts titles, URLs, and snippets from results
Optional content fetching via Jina API (HTML to Markdown)
Parallel fetching of multiple URLs
Configurable maximum number of search results (default 3, max 10)
Graceful error handling for timeouts and failures

Use cases of DuckDuckGo Web Search MCP Server

Search the web and retrieve structured results for analysis
Fetch and summarize the content of specific URLs found in search results
Integrate live web search capabilities into MCP‑powered AI assistants

FAQ from DuckDuckGo Web Search MCP Server

What tools are provided by the server?

Two tools: search_and_fetch (searches DuckDuckGo and optionally fetches content) and fetch (fetches content of a given URL). The search_and_fetch tool accepts query (string) and limit (integer, default 3, max 10). The fetch tool accepts url (string).

How do I configure the server for Claude Desktop?

Add the server to claude_desktop_config.json using either the git+https method or, if that does not work, a local directory path as shown in the README. The command is uvx with arguments pointing to the repository or local folder.

What is the maximum number of search results I can request?

The limit parameter in search_and_fetch has a maximum value of 10. The default is 3.

Does the server require any API keys?

The README does not mention any API keys. The Jina API used for content fetching may have its own terms, but the server itself does not require additional authentication.

What transport protocol does the server use?

The server is MCP‑compliant and is intended to be launched via the uvx command. The README does not specify the transport layer, but typical MCP servers use stdio or HTTP.

Frequently asked questions

What tools are provided by the server?

Two tools: `search_and_fetch` (searches DuckDuckGo and optionally fetches content) and `fetch` (fetches content of a given URL). The `search_and_fetch` tool accepts `query` (string) and `limit` (integer, default 3, max 10). The `fetch` tool accepts `url` (string).

How do I configure the server for Claude Desktop?

Add the server to `claude_desktop_config.json` using either the `git+https` method or, if that does not work, a local directory path as shown in the README. The command is `uvx` with arguments pointing to the repository or local folder.

What is the maximum number of search results I can request?

The `limit` parameter in `search_and_fetch` has a maximum value of 10. The default is 3.

Does the server require any API keys?

The README does not mention any API keys. The Jina API used for content fetching may have its own terms, but the server itself does not require additional authentication.

What transport protocol does the server use?

The server is MCP‑compliant and is intended to be launched via the `uvx` command. The README does not specify the transport layer, but typical MCP servers use stdio or HTTP.