In-depth architectural comparison of the Agent Device and Desktop Touch MCP MCP servers. Compare execution transports, security boundaries, tool capabilities, quality scores, and ready-to-paste client installation snippets for Claude, Cursor, Windsurf, and VS Code.
At a Glance & Executive Verdict
Agent Device
OS Automation · Local stdio
Quality: 52/100 (Good) | Auth: No auth required
Desktop Touch MCP
OS Automation · Local stdio
Quality: 60/100 (Good) | Auth: No auth required
Verdict Summary: Choose Agent Device if you need specialized OS Automation tools running via a local process. Choose Desktop Touch MCP if your workspace requires OS Automation integration with local subprocess execution. Both servers can be configured concurrently in your client's mcpServers manifest.
Which MCP Server Should You Choose?
Choose Agent Device when:
You need dedicated capabilities in the OS Automation domain.
You prefer local stdio subprocess transport architecture.
Your security boundary fits: No auth required (Free / Open Source).
Primary tools included: Cross-platform support for iOS, Android, tvOS, macOS, Linux, and web, Accessibility tree snapshots with token-efficient element references, UI actions including press, fill, scroll, wait, and alert handling.
Discovery router for the agent-device CLI. Exposes status, install guidance, version-matched help, prompts, and resources for iOS, Android, tvOS, macOS, and Linux automation workflows.
Windows desktop automation for LLM agents with entity-based actions instead of coordinate-only clicking. Uses UIA, CDP, screenshots, keyboard/mouse/clipboard, and terminal control, plus entity leases, verified delivery, causal context, and interaction memory to reduce silent UI automation failures.
Category & Scope
Tools & Capabilities Breakdown
Agent Device Tools (5)
Cross-platform support for iOS, Android, tvOS, macOS, Linux, and web
Accessibility tree snapshots with token-efficient element references
UI actions including press, fill, scroll, wait, and alert handling
Evidence capture: screenshots, video, logs, traces, and network data
Replayable workflows with script export and integration with agent terminals
Desktop Touch MCP Tools (23)
desktop_discover
Observe the desktop. Returns interactive entities with leases (UIA, CDP, Terminal, Visual SoM).
Ready-to-Paste Client Configurations
Paste either (or both) of these JSON server blocks into your client config file (e.g. claude_desktop_config.json or ~/.cursor/mcp.json).
Agent Device is categorized under OS Automation and uses a local stdio subprocess. In contrast, Desktop Touch MCP belongs to OS Automation using local stdio subprocess. Select Agent Device when you need capabilities focused on os automation and Desktop Touch MCP when you require tools for os automation.
Perform actions (click, type, drag, select) on entities via lease validation. Returns semantic diffs — plus an optional `roiCapture` (changed-region PNG + next-target preview) on visual-only targets.
desktop_state
Lightweight check of focus, active window, cursor, and Auto-Perception attention signal.
screenshot
Multi-mode capture: `detail='text'` (UIA/OCR), `diffMode` (P-frame), `dotByDot` (1:1), and `background`. Returns a cheap `screenshot://by-ref/{id}` link to the saved image instead of inlining pixels every time.
screenshot_query
Inspect and prune the on-disk screenshot cache behind the by-ref links: `screenshot_query` lists saved captures without re-reading pixels; `screenshot_gc` reclaims space by retention policy (dry-run by default).
workspace_snapshot
Instant session orientation: all window thumbnails + UI summaries in one call.
server_status
Diagnostic check for native engine health and feature activation.