TouchBridge
TouchBridge is a local-first MCP server for iOS simulators and physical devices, enabling AI to see, control, test, and redesign iPhone screens through a structured screenshot bridge and tool surface.
README
<div align="center"> <img src="./assets/touchbridge-mark.svg" width="108" alt="TouchBridge logo" /> <h1>TouchBridge</h1> <p><strong>Let an AI see, control, test, and redesign what is on an iPhone.</strong></p> </div>
TouchBridge is a local-first MCP server for iOS simulators and physical devices. It gives ChatGPT, Codex, Claude, Cursor, and other MCP hosts a reliable action layer for app testing—and a structured screenshot bridge for voice-driven design work.
The design workflow is the unusual bit: “capture this screen for editing” produces a full-resolution image, redacted accessibility geometry, normalized editor layers, the sanitized spoken instruction, a preview, and a versioned manifest. An orchestrating agent can reconstruct the screen in Figma without reverse-engineering the raw accessibility tree.
Why TouchBridge
- One tool surface for simulators and real iPhones.
- Active-device filtering avoids stale, paired, or disconnected hardware.
- Sensitive text is redacted from tool results, logs, and design artifacts.
- Durable design snapshots are useful beyond a single agent session.
editor.jsonprovides named layers, normalized frames, interactivity, image scale, and Figma styling hints.- A loopback-only live viewer never binds to the public network.
doctordiagnoses the entire local toolchain in one call.- CI covers Node 20.17 and Node 22, with no production audit findings.
Voice-to-design loop
flowchart LR
V["Voice request"] --> H["ChatGPT / Codex / Claude"]
H --> T["TouchBridge"]
T --> I["iPhone or Simulator"]
T --> S["PNG + redacted UI map + manifest"]
S --> F["Figma / Slides / image editor"]
F --> H
H --> T
MCP servers do not need to call one another or share credentials. The host agent coordinates TouchBridge and the design tool.
Capabilities
| Capability | Simulator | Physical device |
|---|---|---|
| Tap, swipe, type, and press buttons | Yes | Yes |
| Scan interactive elements | Yes | Yes |
| Read the accessibility hierarchy | Yes | Yes |
| Capture temporary screenshots | Yes | Yes |
| Capture durable design snapshots | Yes | Yes |
| List and launch apps | Yes | Yes |
| Live browser viewer | — | Yes |
| Structured environment diagnostics | Yes | Yes |
Requirements
- macOS
- Xcode command-line tools
- Node.js 20.17 or newer
- Homebrew
- Developer Mode and trust pairing for a physical iPhone/iPad
Run from source
git clone https://github.com/MarshallBear1/touchbridge-mcp.git
cd touchbridge-mcp
npm install
npm run check
npm link
touchbridge --setup-all
Restart the MCP host, boot a simulator or connect an unlocked device, then ask:
Run TouchBridge doctor, find my active iPhone, and test the login flow.
Or:
Capture the current screen for editing, reconstruct the interactive regions in Figma,
and label anything that fails accessibility checks.
MCP tools
| Tool | Purpose |
|---|---|
get_execution_context |
Resolve the active simulator/device and viewer state |
doctor |
Diagnose Node, Xcode, Simulator, idb, image tooling, WDA, and targets |
scan_ui |
Return interactive elements with device-point coordinates |
describe_screen |
Return the visible UI/accessibility hierarchy |
device_action |
Perform one tap, swipe, key, button, or text action |
device_actions |
Perform a sequence of actions |
get_screenshot |
Capture a temporary screenshot for visual verification |
capture_design_snapshot |
Create a durable, redacted handoff with voice intent and Figma-ready layers |
list_devices |
List active simulators and physical devices |
launch_app |
Launch an installed app by bundle identifier |
list_apps |
List installed apps |
setup_device |
Prepare WebDriverAgent commands for a physical device |
See the snapshot contract and the orchestration architecture.
Local data and security
TouchBridge stores its state beneath ~/.touchbridge/.
~/.touchbridge/
├── captures/ durable design handoffs
├── python/ isolated idb environment
├── idb-companion/ local companion support
├── wda-build/ WebDriverAgent source and build data
└── bin/ native accessibility helper
- The live viewer listens on
127.0.0.1only. - Capture names are sanitized and cannot escape the capture directory.
- Password, passcode, PIN, OTP, verification-code, secret, and API-key fields are redacted.
- Text input values are never echoed in MCP results or logs.
- TouchBridge stores no Figma, ChatGPT, Claude, or other third-party credentials.
Development
npm install
npm run typecheck
npm test
npm run check
npm pack --dry-run
CI runs the full check on macOS with Node 20.17 and Node 22.
Roadmap
- Semantic element targeting (
tap "Continue"without manual coordinates) - Screenshot diffing and visual-regression baselines
- Session recording and replayable test scripts
- Editable Figma reconstruction with component matching
- Voice interruption, confirmation, and undo policies
- Multi-device test matrices
- Pluggable privacy policies for regulated apps
License and attribution
TouchBridge is MIT licensed. It includes work derived from the MIT-licensed iPhone-mcp project; the required upstream copyright notice is preserved in LICENSE, with provenance in NOTICE.md.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.