wtf is agent-span?

AgentSpan is a gateway that gives AI agents one consistent way to read and search 52 different internet platforms, such as GitHub, YouTube, Reddit, Wikipedia, and Hacker News, instead of each agent needing its own scraper or integration for every site. It sits between the agent and the internet, offering a REST API, a live event stream, a native MCP server with 91 tools for assistants like Claude Code or Cursor, nine software development kits covering languages such as Python, JavaScript, Rust, and Go, a command line tool, and a React dashboard, all built on an async Rust core. The gateway caches results across three tiers so repeated requests are fast and cheap, and it monitors its own backends: if a connection to a platform starts failing, it automatically switches to a working alternative, reinstalls broken command line tools it depends on, and can alert someone if an outage continues. It also watches its own traffic to suggest things like adjusting how long results stay cached or trying a faster backend. The README is explicit about what AgentSpan does not do. It cannot click buttons, log into websites, or fill out forms the way a browser automation tool can, it is not trying to be a general purpose web crawler for arbitrary sites, and it is not offered as a hosted service, only as software you run yourself. To try it, you run a single command to start the gateway locally, then make simple web requests to read a page or search a specific platform, such as Hacker News. It can also be started with Docker, which brings up the API, a web interface, a Redis cache, a PostgreSQL database, and monitoring tools together. For use with AI coding assistants, a single command writes the correct configuration file automatically. The 52 supported platforms are split into ones that work with no setup and ones that require an API key or login cookie you provide yourself. The project is licensed under MIT.