Cursor AI Controls Raspberry Pi via MCP Server

Peter Harris (Hackster username: zero-onetech) published a step-by-step project on June 12, 2026, demonstrating natural-language control of a Raspberry Pi 4 using Cursor AI and the Model Context Protocol (MCP). The project uses Orbit OS, an Android-like embedded Linux platform that includes an MCP Server app in its app store. Installing the app exposes the device's hardware and system capabilities as a local MCP endpoint at http://DEVICE_IP:9999/mcp. Configuration requires only a single JSON entry in Cursor's MCP config file pointing to that endpoint - no SSH, no Docker, no custom scripts, per the Hackster guide.

The build uses a Raspberry Pi 4 (any RAM variant), a Seeed Studio 2-channel SPDT relay HAT (Relay 1=GPIO5, Relay 2=GPIO6, active-low), and a Seeed Studio BMP280 barometer at I2C address 0x77. Software stack: Orbit OS Community Edition (orbit-os.org), the MCP Server app from the Orbit OS app store, and Cursor AI as the client.

Per the Hackster project, the Orbit OS MCP Server surfaces system info and live metrics (CPU, RAM, disk, temperature, uptime), GPIO control (list pins, set direction, read/write state), I2C bus scan and direct byte transfer, app management (install/uninstall .orb packages), power management (reboot, shutdown), and OTA updates. Harris demonstrates querying device info, listing GPIO lines, reading BMP280 temperature and pressure without specifying register addresses, toggling relays, and remotely rebooting the device - all via natural-language prompts in Cursor.

Orbit OS is developed by Zero-OneTech and uses a lightweight native runtime called Gravity RT and a custom .orb package format instead of Docker containers, per official documentation. Supported devices include Raspberry Pi 2-5, Arduino UNO Q, BeagleBone, NVIDIA Jetson, and other ARM64 boards. Hardware capabilities - GPIO, PWM, I2C, SPI, UART, WiFi, Bluetooth - are exposed as authenticated, encrypted APIs. The platform is available as a Community Edition.

This is a community how-to project (Apache-2.0 license), not a vendor announcement or benchmarked study. Sensor readings and GPIO counts reflect the author's specific build. The project illustrates a broader pattern: pairing standardized protocol layers (MCP) with platform-managed OS runtimes to eliminate per-project glue code for embedded hardware control.