Skip to content

DeepSeek Harness

DeepSeek Harness (dsh) can use your real iPhone. The dsh-esp32agent plugin adds the Handsy MCP server and the handsy skill, so the agent takes a screenshot, taps, swipes and types, then looks again to check the result.

It is a plain hardware add-on: no Mac, no Xcode, no Apple Developer account and no jailbreak. Handsy is not affiliated with DeepSeek.

  1. Get a board: Seeed Studio XIAO ESP32S3 and a short USB-C data cable. See the hardware guide.

  2. Flash firmware v0.2.3 or later from the installer’s TestFlight channel (Chrome or Edge). v0.2.2 works over Bluetooth only.

  3. Pair your computer within 3 minutes of plugging the board into the iPhone (Node.js 18+ and ffmpeg):

    Terminal window
    npx -y esp32agent pair
  4. Set up the iPhone: join the board’s Wi-Fi hotspot (npx -y esp32agent net shows its name and password), turn on Control Center → Screen Mirroring → ESP32 Agent, and enable Settings → Accessibility → Touch → AssistiveTouch. To use your home Wi-Fi instead, see hotspot vs home Wi-Fi.

  5. Install the plugin and restart dsh web:

    Terminal window
    dsh plugin --profile web add dsh-esp32agent
  6. Try a small task in dsh:

    Use the esp32agent tools to control my iPhone. Take a screenshot first, then open Settings, go into Wi-Fi, and tell me which network the phone is connected to. Finally press Home.

Check that the plugin is loaded (use your own profile name):

Terminal window
dsh --profile web --dump-config | grep esp32agent

With several boards, set ESP32AGENT_DEVICE to the board id before starting dsh. Connection problems: see Failed to connect and the CLI guide.

dsh does not ask before each MCP action, and these actions happen on a real phone. The handsy skill tells the agent to ask you before anything that spends money, sends a message, calls someone, deletes data or changes security settings, and to hand passwords and one-time codes to you. Watch the first tasks.

Plugin source: integrations/dsh (MIT).