Betting on MCP: Why we chose an emerging protocol over stable APIs

Building with AI is moving fast, but most of it still feels like 'GPT-wrappers.' When we started building Tooling Studio, we had a choice: stick to traditional, stable REST APIs or go all-in on the Model Context Protocol (MCP).

We chose the latter, and it’s been a 'building the tracks while the train is moving' kind of experience.

The 'Why': Traditional APIs are great for data retrieval, but they don't solve the 'context tax.' We wanted to create a system where the AI isn't just a chatbot, but an agent that can see and act within Google Workspace. MCP provides that standardized bridge.

The Reality of the 'Hard Path':
The Docs are Thin: You spend a lot of time in Discord channels and GitHub issues rather than official documentation.
Breaking Changes: Since the protocol is early, a small shift can break your entire 'bridge.'
The Upside: Seeing an AI actually execute a task, like turning a messy transcript into a structured Kanban board, without a single manual 'copy-paste' is a game changer.

If anyone else here is moving away from standard LLM calls and toward agentic protocols like MCP?

We’re actually prepping for a Product Hunt launch in the next week or so. It’s a bit nerve-wracking to put an 'early access' protocol to the test so publicly, but I’m looking forward to the feedback.

If you’ve launched an MCP-based tool recently, I'd love to know what your biggest 'day-of' surprise was.