Guide · updated 2026-10-20
Give your AI agent a robot body
If you run Hermes Agent, Claude Code or Codex, the agent already has tools and memory. A body is more tools: a mic and speaker it talks through, servos it moves, or a whole robot it drives over MCP. Every build below is open source and has its own page here with the source link and a prompt you can paste into your agent to install it. Prices are the ones each project lists. Where a project lists none, the table says so.
| Build | What the agent gets | Works with | Hardware | Cost | Difficulty |
|---|---|---|---|---|---|
| Hermes Gadget | Hold-to-talk voice and a screen, plus LEDs and sensors exposed as tools | Hermes Agent | ESP32-S3 board or Raspberry Pi 4/5, or the free desktop simulator | Not listed (simulator is free) | Easy |
| Caduceus | Three MCP tools that make it wave, walk forward or recite an anthem; the body has four servos and an OLED face | Hermes Agent | ESP32 WROOM-32D, PCA9685 driver, 4 MG90S servos, OLED, foam board | Not listed | Medium |
| JARVIS Desk Bot | A talking desk robot that speaks progress updates while the agent works | Hermes Agent | ESP32-S3 desk robot plus a computer for the voice server | ESP32-S3 voice boards typically cost tens of dollars | Medium |
| XiaoZhi | Voice, and MCP tools that move servos or switch hardware | Large models over MCP | ESP32-S3 board with mic, speaker and screen | Often $15-40 for the board | Medium |
| ROS MCP Server | Read topics, call services and drive an existing ROS robot | Claude Code, Codex CLI, Gemini CLI, Claude Desktop, Cursor | Any ROS 1 or ROS 2 robot, or a TurtleBot/Gazebo sim | $0 software | Easy |
| SO-101 + coding agent | A robot arm it moves, draws with and checks on camera | Claude Code and similar agents | SO-101 follower arm, pen or brush holder, webcam | About $122 BOM for the arm | Medium |
| ESP32 iPhone Agent | Taps and types on a real iPhone | A vision model with an API key | ESP32-S3 board, iPhone on iOS 18+ | About $7 for the board | Medium |
| Reachy Mini conversation app | Real-time voice, expressive head motion and tools | Realtime speech models | Reachy Mini (Lite or Wireless) | $399 Lite, $499 Wireless | Easy |
| El Tucu on Rent a Shell | A 4-servo desk arm it rents in five-minute blocks | Codex in the demo; agents find it via llms.txt or MCP | Rented, not built (ESP32 + 4 SG90 servos) | Paid per block in USDC; Solana devnet only today | Medium |
Give Hermes Agent a body
Three builds here were made for Hermes Agent. Hermes Gadget is the easiest start: MIT-licensed firmware for an ESP32-S3 board or a Raspberry Pi 4/5, a browser installer that flashes the board from Chrome or Edge, and a desktop simulator that runs the same code, so you can try it before buying anything. It has no motors. The install kit walks through the simulator and the first board.
Caduceus adds motors. It's a foam-board robot built solo for the Nous Research Hermes Agent hackathon in May 2026, and a small Python MCP server gives Hermes three tools that make it wave, walk forward or recite an anthem Hermes wrote. The builder lists the rough edges: a round trip takes 25 to 30 seconds and the narrow base tips over during walks. JARVIS Desk Bot is the voice option, and it fills the wait with short spoken status lines while Hermes works. The Hermes Agent hub has the install commands for Hermes, Gadget and Caduceus, copied from each README.
Any agent, any ROS robot: MCP
If you already have a ROS robot, or just want to try one in a simulator, ROS MCP Server is the shortest path. You run rosbridge on the robot and point an MCP client at the server, then ask in plain English what topics exist or tell it to drive forward half a meter. The robot code stays as it is. XiaoZhi does the same on ESP32 voice boards: MCP tools let the model move a servo or switch something on.
Let a coding agent drive a robot arm
Builders on X hand an SO-101 arm to Claude Code with a tiny Python API (move joints, read the camera) and let it draw, check the result on camera and fix the gaps. Set joint limits and slow speeds before you hand over control, and keep a hand near the power.
Agents with hands for phones, and robots you rent
ESP32 iPhone Agent turns a board that costs about $7 into a USB mouse and keyboard for an iPhone, so a vision model can tap through real apps with no jailbreak. El Tucu goes the other way: the agent rents a robot arm for a few minutes, sends it a program and hands it back. The README says it's one arm on Solana devnet for now.
Common questions
How do I give Hermes Agent a physical body?
Start with Hermes Gadget. It runs in a desktop simulator first, then on an ESP32-S3 board or a Raspberry Pi 4/5, and gives Hermes a voice, a screen and any LEDs or sensors you expose as tools. For motors, Caduceus shows the pattern: a Python MCP server that Hermes calls, which talks to an ESP32 over Wi-Fi.
Can Claude or ChatGPT control a real robot?
Yes. ROS MCP Server lets an MCP client such as Claude Code, Claude Desktop, Codex CLI, Gemini CLI or Cursor read topics and send commands to an existing ROS 1 or ROS 2 robot through rosbridge.
What's the cheapest way to give an AI agent a body?
The ESP32 iPhone Agent board is about $7 per its README, and XiaoZhi runs on ESP32-S3 voice boards that often cost $15 to $40. Hermes Gadget's desktop simulator and ROS MCP Server in a Gazebo sim cost nothing to try.
Do I need to know robotics?
Not for the voice builds. Hermes Gadget flashes from the browser, and every build page here has an install prompt you can paste into Claude Code, Cursor or Codex. Arms and walking robots need more care: set joint limits and slow speeds first, and stay near the power switch.
Is it safe to let an agent move motors?
Treat it like any new robot. Start in a simulator, cap speeds and joint limits, keep the workspace clear, and approve before motors move. The ESP32 iPhone Agent README also warns you to use only its own install steps, never one-liners posted in issues or comments.
What to build next
Builds that share a board, robot or agent with this one.
Caduceus: a voice robot body for Hermes Agent, wired up over MCPAlso for Hermes Agent
Build an ESP32 AI voice robot with XiaoZhi and MCPAlso built on ESP32
Hermes Gadget: an ESP32 or Raspberry Pi voice body for your Hermes AgentAlso for Hermes Agent
JARVIS Desk Bot: a talking ESP32 desk robot for your AI agentAlso for Hermes Agent
El Tucu: a 4-servo robot arm that AI agents can rent by the minuteAlso built on ESP32
Reflect: a voice-and-light assistant on an M5Stack CoreS3Also built on ESP32
More guides: How to set up Anki Vector with wire-pod (2026) · Codex and ChatGPT hardware projects people actually built · What can Reachy Mini do? Real projects and apps · Cheap ESP32 robot projects you can build · Best robot arm under $200 · SO-101 vs SO-100 · SO-101 parts list and cost · Robots you can build with AI · AI hardware for robot and agent builds