# Argent Agentic toolkit that gives AI assistants direct access to iOS Simulators and Android Emulators. ## Agent Decision Summary - Risk level: elevated - Source confidence: high - Recommended workflows: Browser automation, Coding agent workflow, Connector or protocol layer - Permission surface: shell/files, memory, external services - Agent JSON: https://www.openagent.bot/agents/argent.agent.json ## Summary Argent is an agentic toolkit by Software Mansion that connects AI coding assistants directly to iOS Simulators and Android Emulators. It enables agents to tap, swipe, type, launch apps, inspect view hierarchies, read console logs, capture crash reports, profile React Native and Xcode Instruments sessions, and debug mobile apps — all from within the CLI without switching context. It ships as an MCP server and supports Claude Code, Cursor, Copilot, Codex, Gemini, OpenCode, Windsurf, and Zed. ## Guide Mobile simulators are a blind spot for AI coding assistants. An agent can write Swift or Kotlin confidently but cannot launch the app, tap through a flow, or read a crash log — until now. Argent by Software Mansion gives AI agents direct control over iOS Simulators and Android Emulators so they can do what human developers do: build, run, reproduce, debug, and profile. ### What it is Argent is an MCP-native toolkit that connects AI coding assistants to iOS Simulators and Android Emulators. It provides a full suite of capabilities: UI interaction (tap, swipe, pinch, type, gestures, hardware buttons), debugging (console logs, view hierarchy, network requests, React component tree), and profiling (React and Xcode Instruments sessions). It auto-detects editors like Claude Code, Cursor, Codex, Gemini, OpenCode, Windsurf, and Zed, and registers itself as an MCP server at `argent init`. ### Why it matters Mobile development feedback loops are slow. Build → switch to simulator → reproduce → switch back → fix. Argent closes that loop by letting agents operate the simulator directly. Instead of describing what the agent should do, you tell it what you want, and Argent handles reproduction, diagnostics, and profiling. For React Native teams, this is especially powerful — Argent works natively with RN projects so agents can build, launch, and iterate without extra setup. ### How it works Argent runs as an MCP stdio server. After `npx @swmansion/argent init`, your agent discovers Argent's tools automatically. The toolkit talks to simulators directly via `simulator-server` (iOS) and `adb` (Android), bypassing slow XCUITest-based approaches. Every interaction returns structured feedback so the agent knows screen state. For profiling, it runs React and native profiles concurrently and can answer multiple questions from one profiling session without re-running it. ## Use Cases - Autonomous bug reproduction: Tell your agent 'reproduce this crash' and Argent launches the app, navigates through the flow, reads logs, and captures the crash report — all without you touching the simulator. - Performance optimization: Ask your agent to profile a slow screen. Argent records React commit traces and native Xcode Instruments simultaneously, traces the slow commit to a native stack frame, and presents a summary with optimization targets. - Manual QA on simulators: Let agents step through regression flows across iOS and Android simulators, tapping and swiping through screens while logging every interaction. - CI-adjacent debugging: When CI fails with a cryptic simulator error, switch to your agent and ask it to diagnose — Argent attaches the debugger, inspects views, and reads logs in context. ## Alternatives - Choose Argent for mobile simulator control vs browser-based agent tools: Tools like browser-use and OpenClaw cover web automation but do not support mobile simulators. Argent is purpose-built for iOS and Android. - Choose Argent for runtime simulator interaction vs React Native Agent Skills: React Native Agent Skills provides procedural instructions for agents. Argent actually drives the simulator — they work well together. - More reliable than screenshot-based control vs UI automation via screenshots: Argent uses accessibility trees and simulator protocols instead of pixel coordinates, making interactions more reliable across screen sizes and orientations. ### Getting Started - Run the init command: https://github.com/software-mansion/argent - Visit the website: https://argent.swmansion.com ### FAQ - What should I check before using Argent? - You need macOS with Xcode for iOS simulator control, or Android SDK Platform Tools (adb) for Android. Argent runs on Node.js 18+. Review the mixed-license model before production use. - Is Argent open source? - Argent uses a mixed licensing model. Source code is Apache 2.0. Proprietary binaries (simulator-server, ax-service, and native .dylib files) are licensed solely for use within this project. - Which coding assistants does Argent support? - Argent works with Claude Code, Cursor, GitHub Copilot, OpenAI Codex, Google Gemini, OpenCode, Windsurf, and Zed — any agent that can run shell commands and supports MCP. - Does Argent work on Android? - Yes. Argent supports both iOS Simulators and Android Emulators, including UI interaction, debugging, and profiling. - What makes Argent different from XCUITest or Appium? - XCUITest requires switching between test runner and app for every action. Argent talks to the simulator directly, making interactions faster and more suitable for agent loops. It also includes profiling and diagnostics that those tools do not offer. - Is Argent free? - Yes, Argent is free to use and published on npm as @swmansion/argent. ## What It Does Argent is an MCP-native toolkit that connects AI coding assistants to iOS Simulators and Android Emulators. It provides a full suite of capabilities: UI interaction (tap, swipe, pinch, type, gestures, hardware buttons), debugging (console logs, view hierarchy, network requests, React component tree), and profiling (React and Xcode Instruments sessions). It auto-detects editors like Claude Code, Cursor, Codex, Gemini, OpenCode, Windsurf, and Zed, and registers itself as an MCP server at `argent init`. ## How To Evaluate Argent runs as an MCP stdio server. After `npx @swmansion/argent init`, your agent discovers Argent's tools automatically. The toolkit talks to simulators directly via `simulator-server` (iOS) and `adb` (Android), bypassing slow XCUITest-based approaches. Every interaction returns structured feedback so the agent knows screen state. For profiling, it runs React and native profiles concurrently and can answer multiple questions from one profiling session without re-running it. ## Why It Matters Mobile development has a feedback loop problem — build, switch to simulator, reproduce, switch back, fix. Argent closes that loop by giving agents direct control over simulators so they can reproduce issues, inspect state, run performance profiles, and iterate autonomously. For React Native teams especially, this is the first toolkit that lets an agent navigate the app and run diagnostics in the same session. ## Best For - React Native and mobile developers using AI coding assistants - Teams that want agents to reproduce bugs, run UI tests, and profile performance on simulators autonomously - Developers using Claude Code, Cursor, Codex, or Gemini who build for iOS and Android ## Not For - Users who only develop for web or backend and never touch mobile simulators - Teams that cannot install macOS with Xcode (iOS) or Android SDK Platform Tools (Android) ## What It Actually Does - Direct simulator control without XCUITest overhead: Argent talks to the simulator directly instead of routing through XCUITest, making interactions faster and more context-efficient for agents. - Why it matters: XCUITest requires switching between test runner and app for every action, which is slow and wasteful for agent loops. Direct control means the agent can tap, swipe, and type without waiting. - Deep diagnostic suite beyond UI interaction: Argent can attach the debugger, explore view hierarchy, read console logs, evaluate expressions, inspect React component tree, read network requests, and run Xcode Instruments profiling. - Why it matters: Most simulator tools only handle UI interaction. Argent lets agents go from reproduction → diagnosis → fix in one session without leaving the CLI. - React Native and Expo native support: Argent works with React Native apps out of the box — agents can build, launch, and iterate on RN projects the same way they would native apps. - Why it matters: React Native developers are often caught between native tooling and JS workflows. Argent bridges that gap with zero extra setup. - MCP-native architecture: Argent runs as an MCP server over stdio, natively supported by Claude Code, Cursor, Copilot, Codex, Gemini, OpenCode, Windsurf, and Zed. - Why it matters: MCP is the emerging standard for agent-tool communication. Argent requires no custom agent harness — just `argent init` and the agent discovers all capabilities. ## Typical Use Cases - Autonomous bug reproduction on simulators: Ask your agent to reproduce a bug report — Argent launches the app, navigates the UI, reads console logs, and captures crash reports automatically. - Performance profiling and optimization: Agents can record React and native iOS profiles simultaneously, trace slow commits to native stack frames, detect UI hangs and render cascades. - Manual QA automation: Let agents step through test flows, swipe, type, and tap through screens while logging every interaction and screenshot. - CI-adjacent diagnostics: Use in development workflows to let agents diagnose simulator failures, inspect network requests, and read native logs without leaving the terminal. ## How It Compares - Choose Argent for mobile-specific agent workflows vs general-purpose browser agent tools: browser-use and OpenClaw focus on web browser control. Argent is purpose-built for iOS Simulator and Android Emulator control, including deep diagnostics and profiling. - Choose Argent for simulator control vs React Native Agent Skills: React Native Agent Skills provides instructions for agents working on RN projects. Argent is a runtime toolkit that actually controls the simulator — they complement each other. - Deeper than screencap-based approaches vs screenshot-and-click agent tools: Argent uses accessibility trees and direct simulator protocols instead of screen coordinates from screenshots, making interactions more reliable and context-efficient. ## Fit Matrix - Browser automation: strong. Argent has multiple signals for browser automation, including matching tags, capabilities, category, or positioning. Required check: Run one non-sensitive website task and inspect clicks, waits, retries, and changed URLs. - Coding agent workflow: strong. Argent has multiple signals for coding agent workflow, including matching tags, capabilities, category, or positioning. Required check: Run a small repository change and inspect the diff, tests, and rollback path. - Connector or protocol layer: strong. Argent has multiple signals for connector or protocol layer, including matching tags, capabilities, category, or positioning. Required check: Connect one low-risk service, then inspect schemas, auth scope, errors, and logs. - Evaluation and observability: partial. Argent has at least one signal for evaluation and observability, but should be checked against a real task before adoption. Required check: Add one repeatable test case and confirm results can run again in review or CI. - Memory or RAG workflow: partial. Argent has at least one signal for memory or rag workflow, but should be checked against a real task before adoption. Required check: Create, update, retrieve, correct, and delete memory or retrieval objects with real data. - Reusable skill workflow: partial. Argent has at least one signal for reusable skill workflow, but should be checked against a real task before adoption. Required check: Run one skill end to end and check whether it produces evidence or structured output. ## Evidence - verified: Argent is not currently marked as open source in OpenAgent metadata. Source: License metadata: Apache-2.0 with proprietary binaries - verified: Argent has a recorded GitHub repository: software-mansion/argent. Source: Resource facts and GitHub source link. - inferred: Argent supports these recorded deployment modes: cloud. Source: OpenAgent decision signal metadata. - inferred: Argent is tagged with tool calling, mcp, workflow orchestration capabilities. Source: OpenAgent capability taxonomy. ## Missing Checks - Dedicated docs link is missing. - Repository freshness has not been recorded. ## Next Actions - Inspect repository: https://github.com/software-mansion/argent - Open Homepage: https://argent.swmansion.com - Open Source: https://www.npmjs.com/package/@swmansion/argent - Install Argent via npx: npx @swmansion/argent init - Install Argent globally: npm install -g @swmansion/argent && argent init ## Command Line ### Install Argent via npx Run from your project root. The init wizard installs the package globally, detects your editor, and registers the MCP server automatically. ```bash npx @swmansion/argent init ``` ### Install Argent globally Alternative manual install path for global setup. ```bash npm install -g @swmansion/argent && argent init ``` ### Update Argent Pulls the latest version and refreshes workspace configuration. ```bash argent update ``` ### Remove Argent Unregisters the MCP server and uninstalls the package. ```bash argent remove ``` ## Facts - Category: agents - Resource type: agent - Open source: no - License: Apache-2.0 with proprietary binaries - Last verified: 2026-06-03 - GitHub repo: software-mansion/argent - GitHub stars: 3200 ## Capabilities - tool-calling - mcp - workflow-orchestration ## Structured Use Case Tags - developer-workflow ## Getting Started - Open the GitHub repository: https://github.com/software-mansion/argent - Visit the project website: https://argent.swmansion.com - View on npm: https://www.npmjs.com/package/@swmansion/argent ## Links - GitHub: https://github.com/software-mansion/argent - Homepage: https://argent.swmansion.com - Source: https://www.npmjs.com/package/@swmansion/argent ## Structured Outputs - JSON: https://www.openagent.bot/agents/argent.json - Markdown: https://www.openagent.bot/agents/argent.md - Agent JSON: https://www.openagent.bot/agents/argent.agent.json - Canonical: https://www.openagent.bot/agents/argent