Atlas MCP Orchestrator
An internal tool that manages multiple AI coding agents running in parallel. Handles task routing, tracks what each agent is doing, and stops things from falling over when one agent crashes.
Results
The Problem
We were running multiple AI agents to work on different parts of a codebase at the same time, but coordinating them was manual. One person had to watch terminals, assign tasks, check outputs, and restart things when they broke. It was fragile. If the coordinator crashed, everything stopped. There was no visibility into what agents were actually doing beyond reading terminal output.
The Solution
We built a Go-based orchestrator that runs as a background service. It maintains a registry of all active agents in SQLite, handles leader election so there's no single point of failure, and provides a real-time dashboard showing agent status, task progress, and resource usage. Tasks get routed to agents based on availability and capability. When an agent crashes, the orchestrator detects it and reassigns the work. It's not fancy. It does exactly what we need and nothing more.
Implementation Timeline
Design
- Figured out what we actually needed (not what would be cool to build)
- Defined the agent protocol and message format
Core Build
- Control plane in Go
- SQLite agent registry
- Basic web dashboard in React
Reliability
- Leader election so the orchestrator itself doesn't have a single point of failure
- Health checks and automatic agent restart
Hardening
- Load testing with multiple agents
- Documentation
- Deployed to production
“It's not glamorous, but it works. We can run five agents on different parts of a project and actually trust that things won't fall apart when one of them inevitably crashes.”
Related Case Studies
FleetLeaseFlow
A fleet management SaaS that automates invoicing, tracks vehicles in real time, and handles the paperwork that used to eat up the operations team's entire week.
Food Scanner
A mobile app that scans food labels, runs them against the Australian Food Standards Code, and produces compliance reports on the spot. Replaced three-hour paper inspections with twenty-minute digital ones.
Curious where your team stands?
Take the free AI Maturity Map. Benchmark against 5,000+ real-world AI use cases in 5 minutes. No obligation.
Take the free AI assessment