Schema | desktop-automation-mcp

desktop-automation-mcp

Describes the environment variables required to run the server.

Name	Required	Description
`CLICK_EXE`	No	Custom path to the click executable. Defaults to <repo>/tools/CursorClick.exe.
`VISION_MODEL`	No	The vision model ID to use. Aliases: LITTLEAPI_VISION_MODEL, ARK_VISION_MODEL.
`VISION_API_KEY`	No	API key for the vision model. Aliases: LITTLEAPI_KEY, ARK_API_KEY.
`VISION_QUALITY`	No	Image quality setting for vision requests (if supported by the API). Corresponds to config field quality.
`VISION_RETRIES`	No	Number of retries for failed vision requests (default 2). Corresponds to config field retries.
`VISION_BASE_URL`	No	The base URL for the vision model API (OpenAI-compatible /chat/completions endpoint). Aliases: LITTLEAPI_BASE, ARK_BASE.
`VISION_MAX_WIDTH`	No	Maximum width for coarse positioning input image (default 1024). Corresponds to config field maxWidth.
`VISION_MAX_TOKENS`	No	Maximum tokens in the vision API response (default 500). Corresponds to config field maxTokens.
`VISION_TIMEOUT_MS`	No	Timeout in milliseconds for a single vision request (default 90000). Corresponds to config field timeoutMs.
`VISION_DISABLE_THINKING`	No	Whether to disable model thinking (send {"thinking":{"type":"disabled"}}). Accepts 'true'/'false'. Corresponds to config field disableThinking.

Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.

This server publishes no instructions, or was last inspected before Glama recorded them.

Features and capabilities supported by this server

Protocol revision2025-11-25

Capability	Details
`tools`	{}

Functions exposed to the LLM to take actions

Name	Description
desktop_clickA	在屏幕指定物理像素坐标 (x,y) 处执行鼠标单击。高完整性进程注入，带真人式按压时长(默认120ms)。返回是否成功。
desktop_screenshotA	全屏截图并保存为 PNG 文件，返回文件路径。
desktop_locateB	在截图中定位一个 UI 元素/文字/按钮，返回物理像素坐标 (x,y)。采用多阶段精定位：全图校准粗定位 → 局部放大精定位，误差约 ±5px。
desktop_click_targetA	定位目标并点击一步完成：先视觉定位（多阶段精定位），再鼠标单击。返回定位坐标与点击结果。