-
-
instructions transcript
-
tool calls and state log
Flow Control - Web-Native Supervision for Operational Agents
Trying to delegate real-world work to black-box agents using brittle scrapers or custom wrappers is a mess. We need a shared workspace where humans and agents operate on the exact same live state, keeping real authority when things go sideways.
Flow Control turns the web app itself into that control plane using WebMCP.
By moving typed tooling directly to the endpoint, content creators stay in control instead of leaving the world to guess how to scrape an interface.
The app registers authority-aware tools that dynamically shift based on what the human picks—whether it's "Observe," "Assist Me," or "Take the Sector." The agent handles routine traffic using strict, state-version-checked tool calls, but the moment things get exceptional—like a runway blockage—it stops, stages a recovery plan, and pings the supervisor for approval. Both human and agent actions flow through the same engine, logging everything for later audit or review.
While I built this demo around a fictional air traffic control simulation, the architecture isn't locked to ATC. Swap aircraft for warehouse robots, drones, or delivery fleets, and the core pattern stays identical.
Flow Control is an illustrative simulation, not an operational air traffic control product.
Built With
- cloudflare
- react
- typescript
- vite
- webmcp