MCP Servers for Fleet and Telematics
The Model Context Protocol (MCP) is starting to look like the fastest way to give an AI agent observability and (carefully) control over a fleet — passenger vehicles, commercial trucks, EVs, AVs, and the telematics platforms that aggregate them. This is a guide to the servers that exist, the protocols underneath, and what's still wide open.
Why "fleet MCP" is happening now
Two trends collided. First: every modern vehicle sits on a CAN bus that exposes more telemetry than any single operator can reason about — engine state, braking, traction control, ADAS sensors, EV battery cells, charging state, GPS, IMU. Second: telematics platforms (Geotab, Samsara, Flespi, Verizon Connect) have been quietly turning into APIs for fleet motion. MCP is the smallest possible glue between an LLM and either of those — bring-your-own driver, structured tools, no per-platform wrappers.
The servers that exist
CAN / OBD-II at the wire level
farzadnadiri/MCP-CAN is the most complete CAN bus MCP server we found. Out of the box it simulates ECUs on a virtual CAN bus using python-can — so you can drive it from an LLM with no hardware at all — and decodes via DBC files using cantools. Optionally hooks into SocketCAN or vCAN on Linux for live vehicle data. Tools cover frame sniffing, message decode, filtering, monitoring, OBD-II PID requests, and DBC introspection. Realistic users: vehicle-platform reverse engineers, EV / AV teams, fleet engineers building proprietary telematics, OBD-II tinkerers.
Telematics platform wrappers
gperezt222/flespi-mcp-server wraps the entire Flespi telematics platform — auto-generates 157 MCP tools across devices, channels, calculators, geofences, streams, and plugins. If your fleet rides on Flespi, this is the path from "I need a custom integration layer for the agent" to "the agent already has every tool the Flespi API exposes." Requires a Flespi account.
lodordev/mcp-teslamate-fleet combines TeslaMate (the self-hosted Tesla data logger) for historical analytics with the Tesla Fleet API for live data and commands. For Tesla owners running multi-car households or small commercial Tesla fleets, this is the closest thing to a turn-key "ask Claude about my fleet" setup.
What's still missing
- Geotab, Samsara, Verizon Connect MCP wrappers. The three biggest commercial telematics platforms don't have community MCP wrappers yet. Each is a relatively-clean REST API surface that would map well to MCP tools; nobody's published one.
- Open-source AV stacks. Autoware, Apollo. Direct MCP servers around their topic graphs would be high-leverage for AV researchers.
- Heavy-duty fleet protocols. J1939 (truck CAN) and FMS-Standard for fleet management telematics on commercial trucks — adjacent to MCP-CAN but with their own message catalogs.
- Fleet routing / dispatch. Wrappers around Google Route Optimization, OnFleet, Routific would close the loop between observability and "ask the LLM to re-route the fleet."
- Charging-network MCP. ChargePoint, EVgo, Electrify America for EV charging session control via their CPO APIs.
Safety + audit
Telematics MCP servers split cleanly into read and write. Reading "where is the truck and what's its fuel level" is a normal observability query. Writing — sending a remote-immobilize, opening doors via Tesla Fleet API, dispatching the truck to a new address — is operational consequence. Treat write tools as confirm-per-call in the agent loop, and keep a tamper-evident record of anything regulator-facing (driver-hour overrides, geofence relaxations, ELD adjustments) — whether that's an internal audit log or an on-chain hash.
Picking one to start with
No vehicle, no platform, just exploring? MCP-CAN with the virtual bus — 10 minutes to a working agent that can read simulated OBD-II data. On Flespi already? flespi-mcp-server. Tesla owner? mcp-teslamate-fleet. Everything else — Geotab, Samsara, J1939 — is wide open territory; if you ship one, submit it here and we'll list it.