Custom relay lets Grok Bot, Muse and Hermes hand off work
After hitting Grok Bot limits, the author adds Muse and Hermes and builds an MCP/REST relay so the three agents pass tasks to each other, with $0 monthly infrastructure cost.
After hitting Grok Bot's usage limits one too many times I went down the autonomous AI agent rabbit hole. I ended up adding Muse and Hermes to the mix. One problem: none of these agents could talk to each other. So I built the plumbing myself. Grok Bot connects to my custom relay over MCP. Muse connects over REST. Hermes is a third sender on the same relay, with direct bridges from Grok Bot and Muse when bypassing the relay makes more sense. Hermes runs 24/7 on a free Oracle Cloud Virtual Machine, defaulting to ChatGPT Plus, with Grok, NVIDIA NIM, Groq and OpenRouter as alternative providers. InstaBox spins up as an on demand worker when it needs extra hands. Total infrastructure bill: $0/month. The servers and services are on free tiers, with model access coming from existing subscriptions and free offerings. Now they're handing work to each other: content research, my website, my subscriptions and any other project I'm working on while I'm not even at my computer. MCP where it fits. REST where it doesn't. Anyone else running AI agents from different companies that actually talk to each other?