Skip to main content
Glama
Harusame64

desktop-touch-mcp

by Harusame64
OverviewSchemaSourceRelated ServersDiscussionsAdmin
TypeScript
Local

desktop_state

Observes current desktop state to return focused window/element, cursor position, modal flag, and attention signal. Use after actions to confirm UI state and decide if recovery is required.

Instructions

Purpose: Read-only observation of the current desktop state. Returns focused window/element, modal flag, attention signal from Auto Perception. Phase 4 absorbs former get_active_window / get_cursor_position / get_screen_info / get_document_state via include* flags. Details: Always returns: focusedWindow (title, hwnd, processName), focusedElement (name, type, value, automationId), cursorPos {x,y}, cursorOverElement (name, type), cursorOverWindow, hasModal (boolean), pageState ('ready'|'loading'|'dialog'), attention, visibleWindows count. Optional fields (default off): includeCursor:true → cursor {x,y,monitorId} (richer than cursorPos). includeScreen:true → screen {virtualScreen, displays[], displayCount, primaryIndex}. includeDocument:true → document {url, title, readyState, selection, scroll, viewport} via CDP (silently omitted on non-Chromium foreground). includeSessionContext:true (or include:['sessionContext']) → sessionContext {origin, consoleSessionId, sessionLabel, sessionState, ownWinStation} for Terminal Services session classification (ADR-017, observability-only). Chromium: cursorOverElement is null (UIA sparse); focusedElement may fall back to CDP document.activeElement; hints.focusedElementSource reports which path produced the row ('view' = engine-perception latest_focus, 'uia' = direct UIA query, 'cdp' = document.activeElement). Does NOT enumerate descendants — use desktop_discover for actionable entity list and window list. Prefer: Use after each action to confirm state. Cheapest observation tool — cheaper than any screenshot. attention='ok' means safe to proceed; other values require recovery (see suggest[]). Set include* flags only when you need the extra data (each adds one syscall or CDP round-trip). Caveats: Cannot detect non-UIA elements (custom-drawn UIs, game overlays). hasModal only detects modal dialogs exposed via UIA — browser alert/confirm dialogs may not appear here. includeDocument requires browser_open (CDP active); silently omitted otherwise with hints.documentUnavailable. focusedElement.value is the focused field's current text, so a plain credential field's value comes back like any other. Masked fields are withheld on the CDP road by rule; on the UIA road the value is whatever the provider serves and nothing here checks for a masked control — a masked field can arrive as mask characters, one per character of the secret, which nothing here distinguishes from its real value. It is also not always present, and hints.focusedElementValueAbsent says why when it is not: 'view_road_has_no_value' (the perception view is preferred and carries no values at all) or 'masked_on_this_road' (the CDP read dropped a masked field). No hint means nothing was dropped on the road that answered — on the UIA road an absent value can still be a field whose provider will not serve one.

Input Schema

TableJSON Schema
NameRequiredDescriptionDefault
portNoCDP port for includeDocument (default 9222).
tabIdNoOptional CDP tab id for includeDocument; omit for the focused tab.
includeNoOptional response-shape opt-in. `['envelope']` returns the self-documenting envelope (`_version` / `data` / `as_of` / `confidence`). `['raw']` forces raw shape (overrides DESKTOP_TOUCH_ENVELOPE=1 server default). Default behaviour is raw shape (compat with existing clients).
includeCursorNoWhen true, add a richer `cursor` field with monitor index alongside the lightweight `cursorPos`. Phase 4: absorbs former get_cursor_position. Default false.
includeScreenNoWhen true, add a `screen` field with all connected display info (resolution, position, DPI, scale). Phase 4: absorbs former get_screen_info. Default false. Use the displayId values returned here in screenshot / window_dock(action='dock').
includeDocumentNoWhen true, add a `document` field with the focused Chrome tab's url, title, readyState, selection, and scroll position via CDP. Phase 4: absorbs former get_document_state. Default false. Requires browser_open (CDP active); silently omitted on non-Chromium foreground.
includeSessionContextNoWhen true, add a `sessionContext` field with the Terminal Services session classification (origin, consoleSessionId, sessionLabel: 'console'|'rdp'|'other', sessionState: 'active'|'connected'|'disconnected'|'locked'|'unknown', ownWinStation). Default false. Equivalent to `include: ['sessionContext']`. Per ADR-017: observability-only — does not gate input. `sessionState: 'locked'` is a heuristic (active + foreground=null + previous sample within 60s saw a non-null foreground); treat it as a generic input-pause signal — it can also fire on secure-desktop transitions (UAC prompt, Credential UI), where the user-visible state is not strictly 'locked' but input is equally unavailable to this session.

Schema Changelog

Changes observed during successful MCP inspections.

  1. Addedv1.12.0
  2. Removedv1.10.4
  3. Addedv1.9.2
  4. Removedv1.8.0
  5. Addedv1.6.0
  6. Removedv1.5.1
  7. Addedv1.5.0
  8. Removedv1.4.3
  9. Addedv1.4.2
  10. Removedv1.4.0
  11. Changed1 schema field changedv1.1.3
    • addedInput schema / properties / include
      Added value: +{
+  "description": "Optional response-shape opt-in. `['envelope']` returns the self-documenting envelope (`_version` / `data` / `as_of` / `confidence`). `['raw']` forces raw shape (overrides DESKTOP_TOUCH_ENVELOPE=1 server default). Default behaviour is raw shape (compat with existing clients).",
+  "items": {
+    "type": "string"
+  },
+  "type": "array"
+}
  12. Addedv1.0.3

TDQS

A4.8/5.0
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations, the description carries full burden and exceeds expectations. It discloses extensive caveats: non-UIA element detection limits, hasModal only seeing UIA dialogs, CDP fallback behavior, masked-field handling with road-specific details, and the 'attention' signal interpretation. It also explains why focusedElement.value may be absent and the meaning of hints. This level of transparency is exemplary and far beyond typical tool descriptions.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is long but well-structured with Purpose, Details, Prefer, and Caveats sections. Every sentence carries substantive information, and the critical guidance (attention='ok' and prefer-after-action) is front-loaded. It is appropriately detailed for a complex tool with 7 optional parameters, though it could be tightened slightly without losing value.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Given the tool's complexity (7 optional params, no required, no output schema, many edge cases), the description is remarkably complete. It covers all return fields, optional inclusions, failure modes (browser_open requirement, non-Chromium, masked fields), and even hints for troubleshooting. An agent has everything needed to invoke and interpret the tool correctly.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 100%, so baseline is 3. The description adds value beyond the schema: includeCursor is described as 'richer than cursorPos', includeDocument explains silent omission on non-Chromium, and includeSessionContext includes ADR-017 context and a heuristic explanation for 'locked' state. These are meaningful additions that help an agent decide when to enable flags.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description opens with a precise purpose: 'Read-only observation of the current desktop state.' It lists exactly what is returned (focused window, element, cursor, modal flag, etc.) and clearly distinguishes itself from desktop_discover by noting it does not enumerate descendants. The verb 'observe' and resource 'desktop state' are specific and unambiguous.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Explicit guidance says 'Use after each action to confirm state' and notes it is the cheapest observation tool compared to screenshots. It explains when to enable each include* flag and warns that includeDocument requires browser_open. It also compares itself to desktop_discover for actionable entity lists, giving clear when-to-use vs when-not-to-use context.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Deploy Server

Other Tools