How Amazon Device Farm + MCP Server Boost Mobile AI Testing Efficiency by 98%

AI‑driven mobile testing bottlenecks

Mobile testing faces environment inconsistency, device fragmentation, test isolation, and heavy manual intervention, making the testing stage a bottleneck.

Limitations of existing open‑source projects

AutoDroid – controls only local Android devices.

Mobile MCP – provides the Model Context Protocol but lacks cloud infrastructure.

Midscene.js – AI‑driven visual testing that still requires manual device management.

AI‑SDLC gap in mobile testing

AI is mature for requirements analysis, code generation, and code review, yet mobile test verification lacks a closed loop, resulting in reduced development efficiency, insufficient quality assurance, high costs, and slower innovation.

Strategic value of cloud real‑device platforms

Integrating AI with a cloud‑based real‑device service offers a standardized, scalable environment, 24/7 availability, and seamless CI/CD integration, lowering the technical barrier for teams.

Amazon Device Farm Managed Appium Endpoint

Released in November 2025, the Managed Appium Endpoint adds a W3C‑compatible API, a hosted Appium service, HTTPS security, and out‑of‑the‑box integration with WebdriverIO and Selenium.

Device Farm MCP Server

A protocol bridge that connects the Model Context Protocol (MCP) with the Amazon Device Farm API. It adds smart device selection, session management, error handling, provides 22 MCP‑specific tools covering the full test workflow, and enables one‑command start‑up for AI‑driven mobile testing.

Environment preparation and tool installation

# Verify Node.js version (requires v18+)
node --version
# Verify AWS credentials
aws sts get-caller-identity
# List Device Farm projects
aws devicefarm list-projects --region us-west-2

# Register the MCP server with Kiro CLI
kiro-cli mcp add \
  --name "devicefarm" \
  --scope workspace \
  --command "npx" \
  --args "devicefarm-mcp-server" \
  --env "AWS_REGION=us-west-2" \
  --env "AWS_PROFILE=default"

Exploratory test case – Advantage Shopping App

Natural‑language instruction used:

create a pixel device to run Advantage.apk, login with admin/adm1n, if success, generate a report

Device selection

list_devices({ platform: "ANDROID" })
// Selected: Google Pixel 8 (Android 14)
// ARN: arn:aws:devicefarm:us-west-2::device:AC2E189FD1154D05BFCC187783715555

Session creation and APK installation

create_session({
  deviceArn: "arn:aws:devicefarm:us-west-2::device:AC2E189FD1154D05BFCC187783715555",
  platform: "ANDROID",
  sessionName: "Advantage APK Test Session",
  apkPath: "Advantage.apk"
})
// Execution time: 71.64 s (upload + install + session start)

Test actions

mobile_save_screenshot({ path: "step1.png" })
mobile_click_on_screen_at_coordinates({ x: 100, y: 200 })
mobile_type_keys({ text: "admin" })

Efficiency comparison

MCP tool load time: 1.41 s

Session creation + APK install: 71.64 s

Traditional local‑device workflow: ≈2.5 h

The cloud‑native workflow reduces total testing time by 98 % and annual total cost of ownership from $265,000 to $11,000 (‑91 %). Device coverage expands from about 5 devices to over 50 real devices.

Core contributions

Protocol bridge between MCP and Device Farm API.

Automatic device selection, session lifecycle management, and error handling.

22 MCP‑specific tools that cover the complete mobile test flow.

One‑command start‑up for AI‑driven mobile testing.

Future outlook

Planned extensions include AI‑Ops scenarios and deeper automation of test‑to‑deployment pipelines.

Resources

GitHub repository: https://github.com/yoreland/devicefarm-mcp-server