Gemini Deep Research Agent Now Speaks MCP
Google quietly turned its Gemini Deep Research agent into one of the most flexible research APIs on the market. The agent now accepts arbitrary MCP server connections — your private financial data, your internal docs, your custom CRM — alongside Google Search, URL Context, Code Execution and File Search. You can run all of them at once, or shut off the open web entirely and let the agent live exclusively inside your data.
Two flavors. deep-research-preview-04-2026 is the speed version, built for streaming back to a UI. deep-research-max-preview-04-2026 is the slow comprehensive cousin. Both are Gemini 3.1 Pro under the hood and shipped via the Interactions API. The Max variant explicitly targets context gathering and synthesis for downstream agents, which is why it costs more time and tokens.
The killer feature is native chart and infographic generation in-line. The agent doesn't just write a report and stop — it generates HTML and Nano Banana visualizations as part of the output. For the first time you can plug a research agent into a dashboard pipeline without bolting a chart library on top.
Google is openly courting financial data providers — FactSet, S&P Global, and PitchBook are all collaborating on MCP server designs so shared customers can drop their data straight into Deep Research workflows. This is the move OpenAI's Deep Research has been begging for and hasn't shipped. If your agent strategy depends on a research primitive that can talk to your private data over a documented protocol, Gemini just became the default. https://ai.google.dev/gemini-api/docs/deep-research
← Back to all articles
Two flavors. deep-research-preview-04-2026 is the speed version, built for streaming back to a UI. deep-research-max-preview-04-2026 is the slow comprehensive cousin. Both are Gemini 3.1 Pro under the hood and shipped via the Interactions API. The Max variant explicitly targets context gathering and synthesis for downstream agents, which is why it costs more time and tokens.
The killer feature is native chart and infographic generation in-line. The agent doesn't just write a report and stop — it generates HTML and Nano Banana visualizations as part of the output. For the first time you can plug a research agent into a dashboard pipeline without bolting a chart library on top.
Google is openly courting financial data providers — FactSet, S&P Global, and PitchBook are all collaborating on MCP server designs so shared customers can drop their data straight into Deep Research workflows. This is the move OpenAI's Deep Research has been begging for and hasn't shipped. If your agent strategy depends on a research primitive that can talk to your private data over a documented protocol, Gemini just became the default. https://ai.google.dev/gemini-api/docs/deep-research
Comments