Natural Language Access to Vehicle Telemetry via the Model Context Protocol
Valer Bocan · Bianca Gușiță
Record
Abstract
Fleet management systems generate continuous GPS telemetry, yet operators access it through specialized dashboards that handle planned queries well but ad-hoc, cross-vehicle questions poorly. The Model Context Protocol (MCP), introduced by Anthropic in 2024, standardizes how large language models invoke external tools, but its use for IoT telemetry is largely unexplored. This paper describes an architecture that bridges a production GPS tracking system to LLMs through an MCP server exposing four atomic telemetry tools. The Go-based tracking backend receives data from two vehicles over the H02 and TLT2H protocols and stores it in MongoDB behind a REST API; a .NET 9 MCP server translates between this REST surface and the MCP protocol. A structured evaluation over 30 queries against multi-year per-vehicle history shows the LLM autonomously chains multiple tool calls without custom prompting. The server is publicly reachable at https://gpstracker-mcp.dataman.ro/ for reproducibility.
Cite
@inproceedings{bocan2026natural,
author = {Bocan, Valer and Gușiță, Bianca},
title = {Natural Language Access to Vehicle Telemetry via the Model Context Protocol},
booktitle = {International Scientific and Expert Conference Education – Engineering – Economics (ISCEEE 2026)},
year = {2026},
address = {Novi Sad, Serbia},
publisher = {Academy of Applied Studies Novi Sad}
}