native mcp tools are the way to go - frontend is essentially dead.

The future is you having an OpenClaw/Hermes agent connected to these native MCPs and you texting on telegram/whatsapp etc.. with your agent to get information.

I'm already doing this - the future is now!!

▲

edifil 10 hours ago | parent [-]

I love it! I want to do the same, I guess mostly just concerned with hitting limits super quick, is there anyway you can see that info on telegram/whatsapp and route to different models? or are you building that into the tools and just using telegram as conversational

	▲	hackerbone 9 minutes ago \| parent [-]
		So the model route layer should be something that you would have to internally build on the inference POV - like opting for an OpenRouter or creating your own AI Gateway. I've created an ai gateway myself with fallback models - this manages TPM/RPM rate limits and works seamlessly!