Phone Harness 📱

phone-harness · let your agent control your phone.

Connect Claude Code, Codex, or any agent to your real phone. iPhone through the Mac's iPhone Mirroring window, Android over adb from macOS, Linux or Windows, or a Phone Harness Cloud phone — an Android over adb, or an iPhone over HTTPS. No jailbreak, no Xcode, nothing installed on the phone. The agent sees the screen, taps, types, and reads the result.

  ● agent: wants to open Weather
  │
  ● find_text("Weather") → (400, 468)
  │
  ● tap(400, 468) → reads the screen → forecast is up
  ✓ done

Try Phone Harness Cloud → Hosted iPhones and Androids with stealth, real numbers, 2FA, and unlimited devices

Get started by sending the setup prompt to your coding agent.

Demo

Task: "Buy me a Waymo to Delah Coffee from my current location."

https://github.com/user-attachments/assets/80b6d38e-0222-481c-93db-9de60d79247a

Setup

Paste into Claude Code or Codex:

Set up phone-harness for me. Clone https://github.com/ShawnPana/phone-harness into ~/.phone-harness, read `install.md` first, install it so `phone-harness` is a command on my PATH, and register it as an agent skill named phone-harness using `phone-harness skill` as the body. Then read `onboarding.md` and walk me through it.

The agent asks which phone is your default and walks you through the parts that need your hands: pairing iPhone Mirroring and granting Accessibility and Screen Recording, or turning on Android developer options and approving adb. It also offers a cloud Android — the same helpers with nothing to pair, and the first 100 minutes free — so an iPhone user can test on Android too. phone-harness --doctor checks the chain. Details in install.md.

Hermes Agent: hermes plugins install ShawnPana/phone-harness/integrations/hermes adds a phone_exec tool and this skill. See integrations/hermes.

Usage

phone-harness <<'PY'
open_app("Notes")
tap_text("New Note")
type_text("hello from the harness")
print([o["text"] for o in ocr()][:10])
PY

Helpers are pre-imported. SKILL.md is the agent's day-to-day guide; helpers.py is the full list. Task-specific helpers live in agent-workspace/agent_helpers.py in a checkout. A pip or uv install uses ~/.phone-harness/agent-workspace (see install.md), creates it from the packaged defaults when it is missing, and never overwrites an existing one. Delete that folder for a fresh default; the next run recreates it.

No phone on your desk? Rent one

phone-harness cloud login            # once: approve in your browser
phone-harness cloud ls               # phones on the account, and any live session
phone-harness cloud start            # cloud.phone if set, else a temporary Android; 15 min (--for 90s, 15m)
phone-harness cloud start ios        # the account's iPhone, when only one is listed
phone-harness <<'PY'
print(screenshot())
PY
phone-harness cloud stop             # billing stops; a saved Android is written back

cloud ls is the catalog (GET /me → available): each phone's label, platform, and id, with a live session on that row (matched by device id, profile id, or the temporary row). A cloud iPhone session does not echo its catalog id (device is the word "iPhone", and profile is an iphone-… id that matches no catalog row). That session is written on the one catalog row of the same platform and kind whose state is running, when that is the only such row and the only unmatched session of that platform and kind. A session that matches nothing is its own row. cloud start with no name starts this machine's cloud.phone setting if one is set, else what the service gives an unnamed start, a temporary Android. A name is a catalog id or label, or ios / android when only one entry has that platform. A cloud Android is reached over adb, so every helper works on it unchanged. A cloud iPhone has no shell: the helpers POST ops over HTTPS, OCR runs on the host, and driving it is in SKILL.md. phone-harness cloud lists the rest, including watch, open, history, and phone for the saved Android. --session SID (or PHONE_HARNESS_SESSION) pins one rented phone for this process.

Updates

phone-harness update upgrades a uv tool (uv tool upgrade phone-harness) or a git checkout (git pull --ff-only). Any other install prints pip install -U phone-harness and leaves it for you to run. The tool never updates itself. On a terminal, once a day, it prints one line on stderr when PyPI has a newer version. Set PHONE_HARNESS_NO_UPDATE_CHECK=1 to silence that notice.

How it works

iPhone. iPhone Mirroring renders the phone as a Mac window and forwards mouse and keyboard as touches. The harness captures that window, OCRs it with Apple's Vision framework for text with tap-ready coordinates, and posts HID-level events for taps, swipes, and typing.

Android. adb is the transport. screencap is the capture, the phone's accessibility tree is the text source, input is the hands. Works over USB or Wi‑Fi, no window needed.

Cloud iPhone. phone-harness cloud start with that phone's catalog id, label, or ios when it is the only one, drives it over HTTPS from any OS. The host runs OCR; taps use the phone's screen points. No adb and no accessibility tree. Day-to-day steps are in SKILL.md.

Same helpers throughout. phone-harness config set platform ios|android picks the desk default.

Limits

• Unlocking the iPhone pauses mirroring; a PIN-locked Android needs the user.
• OCR sees text, not icons. Unlabeled controls need a screenshot and a vision-capable model.
• No multi-touch, no camera or Face ID flows. DRM video renders black.
• Connecting the phone is always the user's job.

Sponsor

phone-harness is free and maintained in my own time. Sponsoring keeps it that way.