Google turns Deep Research into an MCP-native agent for finance and life sciences

On April 21, Google moved Deep Research Max beyond the usual "find me links" assistant pattern. The new mode runs on Gemini 3.1 Pro and is designed to plan a research task, search the web, consult remote MCP servers, read uploaded files and stream back a running summary of its reasoning. That matters less as a chatbot upgrade than as a workflow change. Finance teams, biotech researchers and corporate strategy groups do not need another text box. They need one system that can combine live web evidence, licensed data and internal documents without sending people across five different tools.

Google says Deep Research now works with Google Search, URL Context, Code Execution and File Search, and it can ground its answers in PDFs, CSVs, images, audio and video. The output is more operational than earlier versions too: charts and infographics are part of the workflow. Google also draws a clear line between the regular mode and Max. Max is positioned as the version that consults more sources and catches nuances the December release sometimes missed. That is the real product signal. Google is not just claiming better quality in the abstract; it is telling buyers that deeper investigation will consume more inference and more retrieval, and that some users will pay for that trade.

The most important addition is MCP support. By connecting Deep Research to remote Model Context Protocol servers, Google is arguing that research agents should not stop at the public web. They should also reach into enterprise and licensed systems. The partner list makes that strategy concrete: FactSet, S&P Global and PitchBook are all named in the launch materials. That points directly at the first users Google wants to win over: investors, corporate development teams, consultants and other knowledge workers whose value depends on source coverage and evidence quality.

Public preview in paid Gemini API tiers matters for the same reason. This is no longer only a feature inside the consumer-facing Gemini app. It is becoming something product teams can embed into their own vertical tools. The next thing to watch is whether Google can make remote MCP access reliable enough for daily work, expand its licensed data partnerships and prove that Deep Research Max saves real investigation time instead of merely producing longer answers.