project · 2025-2026
Two MCP servers, one design pattern
Two production Model Context Protocol servers giving AI agents structured access to TomTom APIs. Both open source, both authored end-to-end (TypeScript, design + impl + docs).
Two production Model Context Protocol servers giving AI agents structured access to TomTom APIs. Both open source, both authored end-to-end (TypeScript, design + impl + docs).
See it work (scripted demo)
Press play. The agent receives a plain-language request, picks the right MCP tool to invoke, sends a structured tool-call, and synthesises a grounded response from the result. Hardcoded transcript, but the tool names and argument shapes mirror the real tomtom-mcp server.
The agent never invented a cafe and never guessed a distance. Every claim in the verdict is backed by a structured tool result. That separation is the whole point of MCP: the LLM does narrative, the server does facts.
And the same pattern, applied to traffic analytics
The traffic-analytics MCP exposes the same shape of tools, just over a different API surface. Below is a different scenario, same flow.
Notice how an agent uses the same protocol across two completely different domains: search-and-routing on one server, traffic-anomaly investigation on the other. That portability is what MCP enables.
tomtom-mcp
General-purpose MCP for location, search, geocoding, and routing. First TomTom MCP server, coordinated with Microsoft for Azure Marketplace publication.
- 12+ tools spanning location lookup, places search, route calculation, traffic flow.
- Hardened argument validation, structured tool-call error handling, configurable rate limits.
- Repo: tomtom-international/tomtom-maps-mcp.
tomtom-traffic-analytics-mcp
Specialised MCP exposing TomTom’s MOVE traffic-analytics services for real-time and historical reasoning over road networks.
- 8 tools across area analytics, junction live + archive, route monitoring, traffic flow, live incidents.
- Designed so an agent can reason from “investigate congestion in this corridor” → query plan → tool calls → narrative response without prompt engineering on the user’s side.
- Repo: tomtom-international/tomtom-traffic-analytics-mcp.
The shared design pattern
Both servers share the same opinionated layout:
- API client first, hand-rolled, fully typed, no auto-generated bloat.
- Tools as thin wrappers over the client. Each tool is a single function with a schema for argument validation.
- Error envelopes so agents can recover gracefully when an API call fails.
- Streamable HTTP transport for proper multi-step agent runs.