AI ENGINEERING
IN USE
Production APIs as agent tools
The Tookan API surface converted into Python MCP tools, so LLM agents can act on real dispatch workflows through scoped, guarded endpoints.
- Python
- MCP
- Node.js
- Blast radius
- Live dispatch
Blast radius
The idea
The Tookan API surface already encoded years of dispatch logic. Converting it into Python MCP tools meant an LLM agent could act on that logic directly — creating tasks, assigning agents, reading fleet state — through the same endpoints the platform uses, rather than through a scraped interface.
AGENT → PRODUCTION
Agent decides what to do, using tools rather than scraped screens.
Engineering
- WHY TOOLS, NOT SCRAPING
- An agent driving a UI breaks the moment the UI changes. Tools bound to real endpoints fail loudly instead, and can be scoped.
- BLAST RADIUS
- These endpoints dispatch real drivers, so the tool layer carries the guards rather than trusting the caller to be careful.
- VERIFICATION
- Mocha.js integration and workflow tests cover critical Tookan flows, so agent-driven actions can be validated against real business behaviour.
- SHARED CONTEXT
- Paired with the engineering memory layer, an agent picks up organisational context instead of being re-briefed each session.