Why do AI agents need web search?

AI models have a training knowledge cutoff — they don't know about recent events, new products, or pages published after their training data. Web search gives agents access to live information, making them dramatically more useful for research, news, and fact-checking tasks.

Why SearXNG instead of a paid search API?

SearXNG is a free, open-source, self-hosted metasearch engine that aggregates results from Google, Bing, DuckDuckGo, and more — with no API key, no rate limits, and no cost per search. Your queries stay private and never leave your infrastructure.

How do I start SearXNG?

The fastest way is docker compose up -d from the agentic-store-mcp directory. This starts SearXNG on http://localhost:8080. Then set SEARXNG_URL=http://localhost:8080 in your MCP config and restart your AI client.

Does it work without Docker?

SearXNG itself requires Docker (or a manual Python install). The agentic_web_search MCP tool is pure Python and has no Docker dependency — only the SearXNG service it connects to needs to run somewhere.

Can I use an existing SearXNG instance?

Yes. If you already have SearXNG running (locally or remotely), set SEARXNG_URL to its address in your MCP client config. The tool connects to any SearXNG-compatible endpoint.

What search categories are supported?

SearXNG supports general, news, images, science, it, files, and more. Pass a category parameter in the tool call to filter results. Defaults to general.

How is this different from agentic_web_crawl?

agentic_web_search takes a search query and returns a ranked list of results — like a search engine. agentic_web_crawl fetches a specific URL you already know and returns its full content and metadata. Use search to discover pages; use crawl to read them.

Is there a rate limit?

No built-in rate limit from the tool itself. Rate limits depend on what SearXNG can fetch — self-hosted instances query public search engines directly, so very high volume may trigger IP-level rate limits from those engines.

AgenticDataRequires SearXNG

Free Web Search for AI Agents

CLI Tool Name: agentic_web_search

Give your AI agent real-time internet search — for free. agentic_web_search queries the web via a self-hosted SearXNG instance and returns ranked results with title, URL, snippet, and source engine. No API key. No per-search cost. No rate limits imposed by the tool.

Why this exists

The problem

AI agents have a knowledge cutoff. Most web search solutions for agents cost $5–50/month or charge per query. APIs like Bing Search, Serper, and Brave all require paid keys.

The solution

SearXNG is a free, open-source metasearch engine you run yourself. Pair it with this MCP tool and your agent gets full internet search — completely free, completely private.

Get web search running in 2 commands

SearXNG runs as a Docker container. Once it's up, set the URL in your MCP config.

bash

# 1. Start SearXNG (one time)
git clone https://github.com/agenticstore/agentic-store-mcp
cd agentic-store-mcp && docker compose up -d

# 2. Run the MCP server with web search enabled
SEARXNG_URL=http://localhost:8080 uvx --refresh agentic-store-mcp

Or add it to your AI client config permanently:

claude_desktop_config.json

{
  "mcpServers": {
    "agenticstore": {
      "command": "uvx",
      "args": ["--refresh", "agentic-store-mcp"],
      "env": {
        "SEARXNG_URL": "http://localhost:8080"
      }
    }
  }
}

[tip]

Using Docker for the MCP server? See the Docker config with web search →

Parameters

Parameter	Type	Required	Description
query	string	yes	The search query string — exactly what you'd type into a search engine.
category	string	no	Search category. Options: general, news, images, science, it, files. Defaults to general.
language	string	no	Language filter (e.g. en, de, fr). Defaults to all languages.

What your agent can now do

→Research any topic with current, live web results — not stale training data

→Find recent news, announcements, and events your model doesn't know about

→Look up documentation, package versions, and API changes published after model training

→Discover competitors, products, and companies with real-time web intelligence

→Answer questions that require up-to-date information (prices, events, new releases)

→Cross-reference claims against current sources before responding

Example output

json

{
  "query": "open source MCP tools for AI agents",
  "results": [
    {
      "title": "AgenticStore MCP Tools — Free, Open-Source",
      "url": "https://agenticstore.ai/docs",
      "snippet": "Give your AI agents real capabilities — Python linting, secret scanning, web search. Self-hosted. No account required.",
      "engine": "google",
      "score": 0.98
    },
    {
      "title": "Model Context Protocol — Anthropic",
      "url": "https://modelcontextprotocol.io",
      "snippet": "MCP is an open standard that enables AI models to interact with external tools and data sources.",
      "engine": "bing",
      "score": 0.91
    }
  ],
  "total_results": 8,
  "category": "general"
}

[note]

SearXNG required. This tool will not function unless a SearXNG instance is running and accessible at the URL set in SEARXNG_URL. Run docker compose up -d from the repo root to start it. The agentic_web_crawl tool works immediately with no setup.

Client integrations

Learn how to connect the agentic_web_search tool to your AI agent:

→ Setup for Claude Desktop → Setup for Cursor → Setup for Windsurf → Setup for VS Code

Explore other AgenticStore MCP tools

→ Web Crawl & Page Extractor MCP Tool (no setup required)→ Python Code Linter MCP Tool → Secret & Repo Scanner MCP Tool → Dependency Vulnerability Scanner MCP Tool

← agentic_web_crawl client setup →

Web search FAQ