I was curious. OpenAI’s promise: build agents with MCP integrations, connect my stack (Cloudflare, GitHub, Linear), automate security reviews. 

‍

The reality? A masterclass in what "beta" really means. 

‍

The interface threw me back to 90s automation platforms – you know the ones with nodes and arrows everywhere. Not inherently bad, but the learning curve was steeper than I expected. 

‍

‍

Then came the security concerns. No OAuth 2.1 despite MCP's own spec describing it. Instead: plain-text API keys. For every. Single. Integration. My security-focused use case suddenly felt ironic. 

‍

‍

MCP isn’t well supported. The only MCP server that fits my use case is Cloudflare, but it’s broken. The original Cloudflare MCP server is available in GitHub, and the readme is clear on the auth method: OAuth. Not an API key.

‍

‍

Function builder works well. OpenAI has code generation down well. It’s simple: hit ‘Generate’, type in what I want, and get code. But it lacks basic security. 

‍

‍

It falls apart using variables. There is a feature called “Set state” where I can set variables. I tried securing my credentials here, but without an option for credentials, it’s still plain text. Worse, I kept getting errors when using this feature. 

‍

But here's what really got me: The hallucinations weren't edge cases. They were consistent. Non-existent Linear tickets. GitHub files that never existed. An agent is confidently wrong about verifiable facts. 

‍

Two hours later, still no working agent. 

‍

The takeaway? We're in an awkward phase. The vision is compelling. We are all looking for agents that truly integrate our workflows. But the execution isn't enterprise-ready, and the fact is that their MCP cannot support Enterprise Tech Ops. 

It's a reminder: Innovation doesn't follow a straight line. Sometimes the most valuable hours are the ones that show you the gap between promise and product. Agent builders are everywhere right now, but most stop short of what real operations need. Security, Dev, and IT teams can’t rely on brittle connectors or a patchwork of API glue that breaks the moment a schema changes. The real unlock is a single, callable environment where every tool and action lives in one chat, where automation, context, and control coexist. That’s what Chat Actions in Kindo delivers: secure, verifiable execution across your stack, without the fragility or friction of builder sprawl.

Get a demo of Kindo and start building today.