AI NewsOpen sourceAnnouncement
RustXNG is a self-hosted web search AI agents can call for free
RustXNG is a self-hosted search engine for AI agents. It runs on your own machine and needs no paid search API key.
Image: GitHub
Why it mattersMost agents that search the web pay a per-call fee to a vendor. RustXNG moves that cost to one small server you run yourself.
Most AI agents that answer from the live web call a paid search API like Bing, Serper or Exa. The bill scales with every question an agent asks. RustXNG offers a different trade. Run your own search engine. Call it from any MCP client. Pay only for the small server you host it on.
RustXNG is a single Rust binary built by Sahen, a student developer in Japan. The repository has 977 stars on GitHub and 82 forks. Main has 151 commits, and the project sits at version 0.31.2. The licence is AGPL-3.0. Anyone who runs a modified copy as a public service has to share their changes. Sahen's profile says most of their code is written with AI assistance.
What it does
RustXNG is based on the SearXNG design. SearXNG is a Python project that asks many search engines at once. It merges the results and does not track the user. RustXNG rewrites that idea in Rust. It adds two things an AI agent needs. The first is a JSON API with ranked results. The second is an MCP server that an agent can call directly.
The project has three provider profiles. Lite is one general engine and runs with no setup. Standard adds 34 keyless engines that need no account or paid key. Full adds billed providers like Exa Search for teams that want them. Any engine can be turned on or off per deployment.
How an agent uses it
The README says the MCP server exposes two read-only tools. search returns up to 10 ranked results for a query. read_page fetches the text of a URL that came back from a search in the same session. The agent can read that page and quote it. The README says the engine does not keep a web-scale index of its own. It is a gateway in front of other engines plus a page reader.
Three guardrails sit between the agent and the open internet. The server checks every page fetch against the site's robots.txt rules. It blocks requests to private network addresses so a prompt cannot reach cloud metadata or the home router. And it refuses URLs the agent did not first see in a search result from the same session.
The project is distributed as a Docker image and as release binaries for Linux, macOS and Windows. The README says a Raspberry Pi 4 is enough to host RustXNG for a small team.
What a team gives up
The README is open about the trade. RustXNG does not build its own web index, so it inherits the quality of whichever search engines it calls. A query that reaches a hostile or rate-limited provider returns fewer results. A query that depends on recency is as fresh as the slowest engine in the pool. And an agent that over-calls search still costs compute on the host, even when the per-call bill is zero.
A team who already pays for a vendor search API gets paid-grade relevance. They also get support when it breaks and a cleaner billing line. A team who runs RustXNG gets the code and the server. The pitch is the one SearXNG has made to self-hosters for years. The new part is an MCP server built in.
Source
sahenjp/RustXNG on GitHub. SearXNG for context: searxng.org.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.