Sofya

Sofya is an AI web search and research API tool that provides web search, page fetching, structured data extraction, and deep multi-source research for AI agents and developers. Built around a single API key and credit-based pricing, it exposes four endpoints — Search, Fetch, Extract, and Research — accessible via both REST and MCP, returning clean markdown output rather than raw HTML or snippet fragments.

The Search endpoint returns full page content (not just SERP snippets) at 1–3 credits per query, with a configurable depth to balance speed and detail. Fetch converts any URL — including PDFs and DOCX files — into clean markdown at 1 credit per URL. Extract uses AI to pull specific structured information such as pricing tables, contact details, or product specs from a page for 5 credits. Research breaks a question into sub-queries, searches up to 30 sources, and synthesizes a cited report for 25 credits.

Sofya has published benchmark results against six competing search backends on OpenAI's SimpleQA factuality benchmark. In agentic mode (search-and-reformulate), it scored 97% accuracy — a statistical tie for first place — while using the fewest average searches of any backend tested (1.12 searches per question). In single-shot retrieval it ranked third at 87.3%, ahead of Brave, Linkup, Firecrawl, and Tavily.

The API is compatible with Claude Code, Cursor, Codex, Windsurf, and VS Code, and supports an MCP server interface at sofya.co/SKILL.md. Free tier access grants 1,000 credits per month for GitHub-verified developer accounts; additional credits cost $0.005 each and never expire.

Best for: developers and AI agent builders who need a reliable, benchmarked web-search and research API with honest pricing and full-page content retrieval.

web search for AI agents, page fetching to markdown, structured data extraction, deep multi-source research, developer API integration, agentic pipelines