ardupilot-mavlink-mcp
Enables AI agents to interact with an ArduPilot vehicle in real-time via MAVLink, including reading state, inspecting and changing parameters, switching flight modes, diagnosing arming failures, and gated arming/disarming.
README
ardupilot-mcp
An MCP server that lets an AI agent talk to an ArduPilot vehicle over MAVLink. Read state, inspect and change parameters, switch modes, read prearm failures, and (gated) arm or disarm. SITL-first.
Install: pipx install ardupilot-mavlink-mcp
mcp-name: io.github.rmeadomavic/ardupilot-mavlink-mcp
[!WARNING] This tool can ARM and command a real aircraft. A bad command can spin props or fly a vehicle away. Defaults are built to stop that: actuation is OFF unless you pass
--enable-actuation, and even then it refuses a real (non-loopback) link unless you also pass--allow-real-vehicle. Develop against SITL. On hardware, bench-test with props off first. No warranty — you own the outcome.
Why
Most ArduPilot tooling for LLMs targets post-flight log analysis. This one drives the live link: connect to a running vehicle, read its state and params, change modes, and diagnose why it won't arm — in the moment, not after landing. The useful case: point an agent at a vehicle that won't arm, have it read the params and the prearm STATUSTEXT, and tell you why, instead of you squinting at a GCS message log. The arm tool reports the real COMMAND_ACK result and hands back the prearm reasons on refusal; it never force-arms.
Architecture
agent (MCP client) ardupilot-mcp vehicle
┌──────────────────┐ JSON-RPC ┌──────────────────────┐ MAVLink ┌──────────┐
│ Claude / etc. │ ───stdio────▶ │ FastMCP tools │ ──udp/tcp/ │ ArduPilot│
│ │ ◀─────────── │ │ │ serial──▶ │ (SITL │
└──────────────────┘ │ ▼ │ ◀────────── │ or FC) │
│ recv thread (1 reader) └──────────┘
│ ├─▶ message cache (latest/type)
│ ├─▶ param store (request/collect)
│ └─▶ COMMAND_ACK + STATUSTEXT
└──────────────────────┘
MAVLink is an async stream; MCP tools are synchronous. One background thread owns the link and is the only reader — it caches the latest message of each type and routes PARAM_VALUE into a param store. Tool calls read from those caches (params block until the data arrives). No two threads ever call recv_match.
What it does
- Reads vehicle state from cache, instantly. Mode, armed, GPS fix and sats, battery, attitude, position — served from cached telemetry, no blocking on the link.
- Diagnoses a no-arm.
ardupilot_armreturns theCOMMAND_ACKresult and, on refusal, the prearmSTATUSTEXT(e.g.AHRS: waiting for home,Accels inconsistent). Safety checks are respected — noARMING_CHECK=0, no force-arm magic number. - Gets and sets parameters on the real param table, with
setconfirmed by the echoedPARAM_VALUE. - Switches flight modes by name, mapped per vehicle type (Copter/Rover/Plane/Sub) — not hardcoded numbers.
Quick start (SITL)
You need an ArduPilot SITL instance. From an ardupilot checkout:
# starts ArduCopter SITL; serves MAVLink on tcp:127.0.0.1:5760
sim_vehicle.py -v ArduCopter --console
Install and run the server (read-only by default):
pipx install ardupilot-mavlink-mcp # or: uv tool install ardupilot-mavlink-mcp
ardupilot-mavlink-mcp --connect tcp:127.0.0.1:5760
To allow arming against SITL, add --enable-actuation.
Use with an MCP client
Claude Code:
claude mcp add ardupilot -- ardupilot-mavlink-mcp --connect tcp:127.0.0.1:5760
Claude Desktop / any mcpServers config:
{
"mcpServers": {
"ardupilot": {
"command": "ardupilot-mavlink-mcp",
"args": ["--connect", "tcp:127.0.0.1:5760"]
}
}
}
Connection strings are pymavlink syntax: tcp:127.0.0.1:5760 (SITL), udp:127.0.0.1:14550, serial:/dev/ttyACM0:115200.
Tools
| Tool | Kind | What it does |
|---|---|---|
ardupilot_connect |
— | Connect to a vehicle. Default is local SITL. |
ardupilot_vehicle_state |
read | Mode, armed, GPS, battery, attitude, position — from cache. |
ardupilot_recent_statustext |
read | Recent STATUSTEXT/prearm messages. Read this to see why arming failed. |
ardupilot_get_param |
read | Read one parameter. |
ardupilot_set_param |
write | Set a parameter, confirmed via echoed PARAM_VALUE. |
ardupilot_list_params |
read | List params, optional glob (ATC_RAT_*). |
ardupilot_set_mode |
write | Set flight mode by name. |
ardupilot_arm / ardupilot_disarm |
write | Gated. Confirmed via COMMAND_ACK. |
Write tools are gated and carry the MCP destructiveHint. Live telemetry is also exposed as the resource ardupilot://telemetry.
Supported vehicles
| Vehicle | Firmware | Status |
|---|---|---|
| ArduCopter | 4.5 | ✓ validated on SITL |
| ArduRover (UGV/USV) | 4.x | ~ mode map present, not yet validated |
| ArduPlane | 4.x | ~ mode map present, not yet validated |
| ArduSub | 4.x | ~ untested |
MAVLink2 is assumed. Not flown on hardware — SITL only so far.
Safety model
- Actuation tools are OFF by default. Enable with
--enable-actuation. - Even enabled, actuation on a real (non-loopback/serial) link is refused unless
--allow-real-vehicleis also set. - Link classification fails safe: anything not clearly loopback is treated as a real vehicle.
- Arming respects the vehicle's safety checks. There is no force-arm option and no tool to disable
ARMING_CHECK.
Status
Working: read, param, mode, and gated arm/disarm against ArduCopter SITL. Validated three ways — unit tests (Python 3.10–3.12), a real-MAVLink-wire check (scripts/wire_check.py), and a live ArduPilot SITL run (scripts/sitl_check.py).
Open targets: validate Rover/Plane/Sub; mission upload/download and guided flight (takeoff/goto/land) are deferred — mission protocol is a stateful handshake and guided commands are fly-away risk. See ROADMAP.md.
Develop
git clone https://github.com/rmeadomavic/ardupilot-mcp && cd ardupilot-mcp
uv venv && uv pip install -e ".[dev]"
pytest -q
ruff check . && ruff format --check .
python scripts/wire_check.py # offline real-wire check, no SITL needed
License
MIT — Kyle Adomavicius
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.