Yeah, limitations like this are a real buzz-kill. @Bharat_Batra1 is hopefully watching.
I overcome these limitations in most use cases with two [agentic] rules baked into mcpOS:
- Authorize the agent to use a browser subagent.
- Instruct the subagent to verify and thoroughly test its approach in a document “sandbox” before implementing in the production document.
Here’s the general process I’ve defined in my agentic platforms, Antigravity and Claude Code/Cowork:
It’s important to note that browser subagents are not perfect, but in most cases, they work well. Given their somewhat stochastic nature, ambiguity can also be a buzz kill. This is why it’s so important to add a clarity layer to any agent that is writing content to a Coda document. Coda MCP alone is insufficient (and technically unable) to provide this layer as I exposed in this agentic strategy article.
