SearXNG AI Kit MCP
SearXNG AI Kit gives agents web search across 180+ engines through a CLI and an MCP server, with no search API key
- 110 ★ on the repository
- Last commit
Extended access
Why this level
- Sends search queries to external search engines, and page reading goes through Jina.ai
- The `ask` tool calls a language model and spends your account's tokens
- The optional CLI Proxy API link uses provider subscription logins, so check it against the vendor's terms
Install
- Works in
Manual install
{"mcpServers":{"searxng":{"command":"searxng","args":["mcp-server"]}}}The `searxng` binary must be on the PATH.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Sends search queries to external search engines, and page reading goes through Jina.ai
- The `ask` tool calls a language model and spends your account's tokens
- The optional CLI Proxy API link uses provider subscription logins, so check it against the vendor's terms
This is third-party code. Review the repository files before installing.
What it does
SearXNG AI Kit packs the SearXNG metasearch engine into a single `searxng` program with a command-line search, a Python library and an MCP server. It queries search engines directly, so no paid search API key is needed; `searxng engines` lists engines such as DuckDuckGo, Startpage, Brave, Google and Bing. `searxng search "query"` searches with defaults, and the flags `--engines`, `--category`, `--lang`, `--time` and `--format json` narrow the results. `searxng fetch-urls` downloads and cleans the text of several pages in parallel. The `searxng ask` mode and the interactive `searxng chat` answer a question by calling search and page reading themselves through a chosen language model. `searxng mcp-server` starts an MCP server with five tools: `web_search`, `multi_web_search`, `fetch_url`, `fetch_urls` and `ask`. Pages are read through Jina.ai, and `JINA_API_KEY` is optional. Ready binaries for Linux and macOS are on the releases page.
Who it is for. Developers who want to give Claude Desktop, OpenCode or another agent web search without a paid API and without sending every query to one search company.
Good fit when
- The agent needs web search across several engines at once and you have no Brave or Tavily key
- Research needs a dozen parallel queries and `multi_web_search` fits
- You need the text of several links in one call with result order preserved
- A script needs JSON results from the terminal: `searxng search "query" --format json`
- You want search in a Python project as a library, without running your own SearXNG server
Not a fit when
- You work on Windows, where no ready binaries exist
- You need search that never fails: engines sometimes block requests with CAPTCHAs or rate limits
- You want the full SearXNG web interface for people
- The agent's built-in search is enough, for example in Claude or ChatGPT
Example request
Search with searxng for three recent reviews of Valkey, fetch their text and compare the conclusionsMCP
- Transport
- stdio
- Authentication
- not required
Environment variables
JINA_API_KEYsecret- Jina.ai key for extended page reading
Limitations
- Ready binaries are published for Linux and macOS, and they are disabled for Windows for now. Building from source is slow because it builds a pinned SearXNG version.
- Results depend on engine availability. CAPTCHAs and rate limits can block some engines, and errors go to stderr.
- The author warns that the built executable can have engine loading problems and recommends the Python version.
- The `ask` tool uses an OpenAI o3 model by default and needs access to a language model. Page reading goes through the third-party Jina.ai service.
How to disable. Remove the `searxng` entry from MCP settings, for example from `claude_desktop_config.json`, and delete the binary. If you installed through uv, `uv tool uninstall` removes it.
FAQ
How do I connect SearXNG AI Kit to Claude Desktop?
Add a `searxng` server to `claude_desktop_config.json` with the command `searxng` and the arguments `["mcp-server"]`. The binary must be on the PATH first.
Is a search API key required?
No, queries go to the engines directly. Only `JINA_API_KEY` is optional, for extended page reading.
Which tools does the MCP server expose?
Five: `web_search`, `multi_web_search`, `fetch_url`, `fetch_urls` and `ask`. The last one calls a language model that uses the other four itself.
Why does search sometimes return nothing?
Some engines respond with an error, a CAPTCHA or a rate limit. Try a narrow set such as `--engines duckduckgo,startpage` and look at the stderr output.
Does SearXNG AI Kit work on Windows?
Windows binaries are disabled until the package launcher becomes native. Binaries for Linux and macOS are on the releases page.
Discussion
No questions or reviews yet.
Related
MCP serversAn MCP server for private web search through your SearXNG instance: search with pagination and filters, page and PDF reading
MCP server for web search and crawling: SearXNG and You.com plus clean Markdown from pages, sites and logged-in areas
Local web search for agents: searches, reads pages and returns compact passages with sources, saving context tokens
A deep research MCP server: multi-source search, result fusion and cited synthesis through a Qwen3 model on OpenRouter
A local MCP server with no API keys: multi-engine web search, page fetching and document reading tuned for LLMs.
Searcharvester runs your own search for agents: SearXNG, page extraction to markdown and a team for deep research.
README in short
The README sets out the kit's features, installing from a binary, through Nix and from source, example search and engine commands, the AI assistant mode and model setup. The last large section covers the MCP server and its tools.