Firecrawl edges out Browserbase's WebMCP in a week of agent-first tooling

Agent-native search and page interaction

The week's most significant release belongs to Firecrawl, which launched its Research Index, a specialized search index built for AI/ML literature and the code that powers it. The index covers 3M+ arXiv papers alongside GitHub artifacts — issues, merged PRs, and READMEs, refreshed daily — and it's already benchmark-leading: 53.3% recall on arXivQA versus 45.4% for the next best provider, with a 0.750 MRR that puts the right paper in the top one or two results. For any agentic workflow that needs to ground itself in recent research or verify claims against full text, this is a material improvement at comparable cost. The index is available via the Firecrawl API, SDKs, MCP, and CLI — and keyless access was also added for core endpoints, lowering the friction to try it.

Browserbase's Stagehand took a complementary step with WebMCP support in version 3.6.0 (also covered in a dedicated changelog entry). WebMCP lets a page expose typed, first-party tools directly to an agent, so Stagehand calls those tools instead of inferring actions from the DOM. Two new methods — page.listWebMCPTools() and page.invokeWebMCPTool() — give developers a clean API for discovering and invoking page-level capabilities. This is a meaningful shift for any agent that needs to interact with complex web UIs reliably: instead of guessing at selectors and hoping the DOM structure holds, the page declares exactly what it can do. The release also adds Azure OpenAI auth via Microsoft Entra ID, support for claude-fable-5 with adaptive thinking and the new xhigh effort, and smarter act caching that normalizes URLs differing only in query-param order.

What ties them together

Both releases are responses to the same problem: agents need better, more structured ways to find and interact with information on the web. Firecrawl solves the search side — a purpose-built index that outpaces general-purpose providers on research recall — and Stagehand solves the interaction side, letting pages declare their own tool surface instead of making agents reverse-engineer the DOM. Together, they make the browser automation stack more capable of supporting agentic workflows end-to-end, from literature discovery to live-page interaction.