MCP Searxng

Overview of MCP Searxng

What is MCP Searxng about? MCP Searxng delivers a server that connects AI agents to a SearXNG meta‑search instance. It forwards search queries, fetches result pages, converts them to Markdown, and streams the data via Server‑Sent Events (SSE), enabling agents to obtain fresh, privacy‑controlled web information.

How to use MCP Searxng?

1. Install the runtime – Install uv (or use Docker) on an Ubuntu 24.04 host.

   curl -LsSf https://astral.sh/uv/install.sh | sh
   

2. Clone the repository

   git clone https://github.com/erhwenkuo/mcp-searxng.git
   cd mcp-searxng
   uv sync
   

3. Start a SearXNG service (Docker‑compose provided)

   cd searxng-docker
   docker compose up -d   # maps http://localhost:8888 to SearXNG
   

4. Launch the MCP server
   • With uv:

     uv run server.py --searxng_url="http://localhost:8888"
     

   • Or with Docker:

     docker build -t mcp-searxng .
     docker run -d -e SEARXNG_URL="http://<host‑ip>:8888" -p 5488:5488 mcp-searxng
     

5. Verify via MCP Inspector (Node.js tool)

   npx @modelcontextprotocol/inspector
   

   Connect to http://localhost:5488/sse, list the tools (web_search, web_url_read), and run them.

Key Features

• SSE transport for cloud‑native, decoupled client‑server interaction.
• Two ready‑to‑use tools: web_search (meta‑search) and web_url_read (page fetch‑and‑markdownify).
• Seamless integration with a self‑hosted SearXNG instance (Docker‑first).
• Configurable host, port, and SearXNG endpoint via command‑line flags.
• Docker images for both the SearXNG backend and the MCP server.

Use Cases

• Enabling AI agents to obtain up‑to‑date information from multiple search engines while preserving privacy.
• Supporting multilingual queries, especially Chinese, where mainstream engines may fall short.
• Building cloud‑native AI workflows that need on‑demand web search without embedding a full browser.
• Rapid prototyping of AI‑driven research assistants that require external knowledge.

FAQ

• Can I run the server on a different port? Yes, use --port <number> when launching the server.
• What if SearXNG runs on a different host? Supply the address via --searxng_url (or the SEARXNG_URL env var for Docker).
• Do I need Docker to use SearXNG? No, you can deploy SearXNG separately; the server only needs the HTTP endpoint.
• Is Python required on the host? Only if you run the server directly (uv). The Docker variant contains all dependencies.
• How do I test the tools? Use the MCP Inspector UI: select SSE, connect, then invoke web_search or web_url_read from the Tools tab.

mcp-searxng

An example of an MCP Server for use by an AI Agent, designed to allow the AI Agent to search for new external information through SearXNG's open-source meta-search engine.

Currently, many search engines other than Google have emerged in the market, attempting to capture market share in areas where Google falls short. For instance, DuckDuckGo emphasizes not tracking users, Ecosia plants trees with every search, and Brave Search aims to harness collective efforts to build a free search engine.

However, the results returned by these engines are often unsatisfactory. Firstly, they don’t crawl as many web pages as Google does; secondly, their support for Chinese is poor. Although they can access some interesting pages that Google doesn’t display, search engines other than Google are still quite difficult to use.

So why not combine the results from multiple search engines!? That’s exactly what a meta-search engine does. SearXNG, an open-source meta-search engine software, can be self-hosted or used via sites provided by enthusiastic community members. For businesses, SearXNG offers a way to maintain privacy and security control while enabling AI Agents to effectively search for the external data they need.

References:

• SearXNG Official Website
• Self-Hosting an Open-Source SearXNG Meta-Search Engine to Search Google, DuckDuckGo, and More at Once

Purpose

This MCP server demonstrates an SSE-based MCP server (integrated with SearXNG and Microsoft's markdownify to extract web pages into Markdown-formatted text) and its operational mode using the MCP Inspector (MCP client).

Runtime Environment

This project uses uv to manage dependencies and the Python runtime environment. If uv is not yet installed, you can follow the installation instructions on the official website.

The following commands are executed in an Ubuntu 24.04 environment. For operations on other operating systems, please adjust accordingly:

$ curl -LsSf https://astral.sh/uv/install.sh | sh

Download source code:

$ git clone https://github.com/erhwenkuo/mcp-searxng.git

$ cd mcp-searxng

$ uv sync

Running the Service

Running the SearXNG Service

First, install Docker on the machine where it will run and perform the related configurations. For detailed information, please refer to: Install Docker Engine on Ubuntu

In the project directory, there is a pre-configured simple SearXNG setup to facilitate testing.

mcp-searxng/searxng-docker/
├── docker-compose.yaml
└── searxng
    ├── settings.yml
    └── uwsgi.ini

Switch to the searxng-docker directory and use Docker Compose to start a SearXNG service:

$ cd searxng-docker
$ docker compose up -d
$ docker compose ps

NAME      IMAGE                              COMMAND                  SERVICE   CREATED          STATUS                    PORTS
searxng   docker.io/searxng/searxng:latest   "/sbin/tini -- /usr/…"   searxng   29 minutes ago   Up 29 minutes (healthy)   0.0.0.0:8888->8080

The test SearXNG service is mapped to the local machine's port: 8888.

Starting the MCP-SEARXNG Service

Method 1. Using uv to start:

Enter the following command to start:

$ uv run server.py --searxng_url="http://localhost:8888"

INFO:     Started server process [219904]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
INFO:     Uvicorn running on http://0.0.0.0:5488 (Press CTRL+C to quit)

Using Docker to Start

First, build the Docker image:

$ docker build -t mcp-searxng .

Start mcp-searxng. Since the mcp-searxng service is being started using Docker, you cannot use localhost to point to the SearXNG service address when configuring the connection to SearXNG. It is recommended to directly query the local machine's IP address and then use the SEARXNG_URL environment variable for configuration.

The startup parameters below assume the local machine's IP is 192.168.54.88:

$ docker run -d -e SEARXNG_URL="http://192.168.54.88:8888" -p 5488:5488 mcp-searxng

Verifying the Results

First, install Node.js:

# Download and install nvm:
curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.2/install.sh | bash

# In lieu of restarting the shell
\. "$HOME/.nvm/nvm.sh"

# Download and install Node.js:
nvm install 22

# Verify the Node.js version:
node -v # Should print "v22.14.0".
nvm current # Should print "v22.14.0".

# Verify npm version:
npm -v # Should print "10.9.2".

Next, start the MCP Inspector:

$ npx @modelcontextprotocol/inspector

Starting MCP inspector...
Proxy server listening on port 3000

🔍 MCP Inspector is up and running at http://localhost:5173 🚀

Open http://localhost:5173 in your browser and perform the following actions:

1. Select SSE in the Transport Type dropdown.
2. Enter the MCP server's address and port in the URL field: http://localhost:5488/sse.
3. Click Connect. If the status shows "Connected," it means you have successfully connected to the MCP server.
4. Click the "Tools" tab at the top.
5. Click the "List Tools" button, and you should see two tools:
   • web_search
   • web_url_read
6. Click web_search. On the right, you’ll see the tool’s description and parameters. Enter the keyword you want to search for in the query input field, then click the "Run Tool" button.

The effect is shown in the image below:

Test web_url_read:

• Click web_url_read. On the right, you’ll see the tool’s description and parameters. Enter the URL of the webpage you want to retrieve in the url input field, then click the "Run Tool" button.

Why Use SSE

This means the MCP server can be a process running remotely, and the AI Agent (client) can connect, use, and disconnect from it anytime, anywhere. In other words, an SSE-based server and client can be decoupled processes (potentially even on decoupled nodes).

Compared to the STDIO-based model, where the client spawns the server as a subprocess, this is different and more suitable for "cloud-native" use cases.

MCP Server

server.py is an SSE-based MCP server. By default, the server runs on 0.0.0.0:5488, but it can be configured using command-line arguments, for example:

uv run server.py --host <your host> --port <your port>

Startup Parameters:

Parameter	Required	Default	Type	Description
--host	No	0.0.0.0	str	Host to bind to
--port	No	5488	int	Port to listen on
--searxng_url	No	http://localhost:8888	str	SearXNG URL to connect to