- First AI Movers Pro
- Posts
- The API Era of Agents—MCP & Mariner Are Your New Secret Weapons
The API Era of Agents—MCP & Mariner Are Your New Secret Weapons
OpenAI’s Model Context Protocol and Google’s Project Mariner bring universal data pipes and browser-level automation—here’s how to build on them before the crowd catches up.
Dr. Hernani Costa
June 27, 2025
Good morning,
Today’s edition is all about agent-ready APIs: OpenAI’s Model Context Protocol (MCP) turns ChatGPT into a plug-and-play data hub, while Google’s Project Mariner gives agents hands-on the actual browser. Together, they reset the playing field for anyone building AI workflows. Let’s unpack what shipped, why it matters, and how you can ship with it—fast.
MCP + Mariner: The Two APIs Every Builder Should Watch
Model Context Protocol—“USB-C for AI”
MCP is an open protocol that standardizes how apps feed tools and data into large language models. Think: OAuth + schema for files, emails, tickets, or even Salesforce records via community connectors, github.com. OpenAI just enabled free Gmail, Drive, Outlook, Dropbox connectors inside ChatGPT—no plug-in install needed.
Builder angle: Instead of writing bespoke RAG pipelines, drop an MCP connector and let ChatGPT search, summarize, or transform data in your SaaS with one prompt. Expect a cottage industry of niche connectors (e.g., Jira sprints, SAP ledgers) by the end of Q3.
Project Mariner—Agents That Click for You
Unveiled at Google I/O 2025, Mariner exposes a Computer Use API letting Gemini agents navigate browsers, fill forms, and juggle up to 10 simultaneous tasks. Early testers (Automation Anywhere, UiPath) are wiring Mariner into RPA flows; broader Gemini API access lands this summer.
Builder angle: Replace brittle Selenium scripts with an LLM that sees the DOM and adapts. Use Mariner’s Teach & Repeat to demo a workflow once—agents will replicate it at scale.
Why both APIs shift strategy
Pain Point | Old Way | New Way |
|---|---|---|
Data silos | Build a custom ETL or RAG for each source | Plug an MCP connector; prompt “Search last Q’s invoices” |
Web automation | Headless browser + XPath headaches | Mariner agent navigates dynamically, retries, explains |
Multi-tool orchestration | Chain scripts & webhooks | One agent calls MCP (data) + Mariner (action) in a loop |
Takeaway: The moat moves to domain depth and workflow UX. If your startup only connects data or clicks buttons, these APIs will undercut you. Innovate on insight, analytics, or compliance on top.
Tool Highlight — Vertex AI Agent Builder
Just announced, Google Cloud’s Agent Builder sits on top of Mariner and Gemini APIs—drag-and-drop UI to orchestrate multi-agent flows without redeploying your stack. Great for internal automation prototypes before you commit code.
Fun Fact—The First Public Web API Was for… a SOAP Bar?
Not quite—but close. Salesforce launched the first enterprise web API in February 2000 so developers could integrate “Sforce” CRM data into apps over XML-RPC (pre-REST). Twenty-five years later, we’re wiring entire browsers into LLMs—proof APIs age like fine wine (with occasional SOAP-y residue).
Wrap-Up & CTA
MCP and Mariner hand you universal data pipes and browser automation. Question: Which will you prototype first—a connector-powered insight bot or a Mariner form-filler? Hit reply and share; your experiments fuel tomorrow’s deep dive.
Until next time—keep prompts sharp and endpoints secure.
What Top Execs Read Before the Market Opens
The Daily Upside was built by investment pros to give execs the intel they need—no fluff, just sharp insights on trends, deals, and strategy. Join 1M+ professionals and subscribe for free.
Reply