Works with Claude, GitHub Copilot, Cursor and any MCP client. Pairs with @testkase/test-management-mcp for test cases, cycles and reports.

Quick start

1. Get an access key: TestKase > API Keys > Create API Key (starts with xyz_).
2. Add it to your assistant (Claude Desktop, Copilot, Cursor: same block in their MCP config file):

{
  "mcpServers": {
    "testkase-device-farm": {
      "command": "npx",
      "args": ["-y", "@testkase/device-farm-mcp"],
      "env": { "TESTKASE_PAT_TOKEN": "xyz_your_token_here" }
    }
  }
}

Claude Code:

claude mcp add testkase-device-farm -e TESTKASE_PAT_TOKEN=xyz_your_token_here -- npx -y @testkase/device-farm-mcp

3. Restart your assistant and ask: "Open saucedemo.com in Chrome and log in", "Why did my last run fail?", "Give me Appium settings for a Pixel 8".

Tools

Drive a device

| Tool | What it does | |------|--------------| | start_device | Open a phone (platform + an uploaded app or a url) or a browser (browser, os, version, optional url; tunnel: true reaches your localhost through Local Tunnel, on Windows browsers). Without a tunnel, local and private addresses (localhost, 10.x, 192.168.x, .local …) are refused; phones always refuse them. | | wait_for_device | Keep waiting when start_device answered starting. | | read_screen | The screen as a numbered list. | | screenshot | A picture of the screen plus the list. | | tap / type_text / swipe / press_key | Act on an item by its number (or an x,y point). | | open_url / app | Go to a page; restart or bring back a phone app. | | run_steps | Up to 10 actions in one call, plus wait_for some text. | | end_device | Give the device back. |

Runs, apps and setup

| Tool | What it does | |------|--------------| | list_devices / get_device | The phone models and browsers on offer (names only). | | list_runs / get_run / list_run_tests | Your Automate runs and their tests. | | get_run_commands / get_run_logs / get_run_artifacts | A run's commands, logs, video and screenshots. | | analyze_failure | Why a browser run failed, in one call. | | stop_run | End a stuck browser run. | | get_connection_config | Hub URL and capabilities for Selenium, Appium or Playwright. | | list_apps / upload_app | Your app builds (.apk, .ipa, simulator .zip). | | get_tunnel_status | Is your Local Tunnel connected? |

How driving works

• Every item on screen gets a number that stays while the item stays. The AI says "tap 5" instead of guessing positions.
• After an action, the answer lists only what changed (new, gone, updated), not the whole screen.
• Browsers: type drop-down choices, dates and colours as text (2026-10-09, #1a2b3c). If a tap by number does nothing, tap its x,y.
• Limits: one device at a time; a session ends after about 3 minutes idle or 60 minutes in total.

Typed text can appear in the run's log and video. Use test accounts.

Your key and privacy

• The key stays on your machine and goes only to TestKase.
• It acts as you: it sees your own runs and apps; the device list is shared in your organisation. Tests you run use your minutes.
• Logs, errors and screenshots you ask about are passed to your AI assistant.

Settings

| Variable | Required | Default | |----------|----------|---------| | TESTKASE_PAT_TOKEN | Yes | | | TESTKASE_HUB_URL | No | https://hub.testkase.com |

Troubleshooting

| Problem | Fix | |---------|-----| | "access key isn't valid or has expired" | Create a new key and update TESTKASE_PAT_TOKEN. | | "TESTKASE_PAT_TOKEN is not set" | Add the env block, then restart your assistant. | | Tools don't show up | Restart the assistant fully; check the JSON in your config. |

Links

TestKase · Docs · @testkase/test-management-mcp

MIT License