Desearch

Desearch is a web search API built for AI agents. One API gives an agent web search, social search across X and Reddit, video search on YouTube, academic search on arXiv, people and company search, and page extraction, so the agent can find sources and read them in the same workflow.

Results come back as structured JSON with titles, links, snippets and page content, which makes them easy to pass into a model's context. Desearch can also return an AI summary of the results, so an agent gets a short answer together with the sources it came from. Three main modes cover most needs: Search for fast lookups, Research for deeper multi source answers, and Extract for pulling clean text out of a given page.

For agent builders, Desearch ships an official MCP server (github.com/Desearch-ai/mcp-desearch), so it plugs into Claude, Cursor and other MCP clients without custom code. There are also SDKs and a REST API for Python and JavaScript projects, plus an onboarding prompt that helps a coding agent set up the integration.

Desearch runs on Bittensor Subnet 22, a decentralized network of miners and validators that compete to deliver good search results. Developers can start with a free API key and move to paid usage as their agents grow. Typical uses include research assistants, market and competitor monitoring, social listening, news tracking, lead and company research, and any agent that needs fresh information from the open web.