Claude KVM
AI 与智能体by aras-workspace
用于通过 VNC 控制远程桌面的 MCP server,基于原生 Swift daemon,并集成 Apple Vision OCR。
什么是 Claude KVM?
用于通过 VNC 控制远程桌面的 MCP server,基于原生 Swift daemon,并集成 Apple Vision OCR。
README
Claude KVM is an MCP tool that controls remote desktop environments over VNC. It consists of a thin JS proxy layer (MCP server) and a platform-native Swift VNC daemon running on your macOS system.
[!TIP] Phantom-WG could be a great alternative for you. Isolate your VNC server within your own network while enjoying self-hosted VPN performance with the extra privacy features you gain along the way.
Live Test Runs
- Integration Test
- Mac Integration Test
- Mac Calculator Test
- Mac Scientific Calculator Test
- Mac Safari Browsing Test
- Mac Drag & Drop Test
- Mac Chess Test
- Mac Chess Direct Test
- Mac Phantom-WG Install Test
[!NOTE] Tests are conducted transparently on GitHub Actions — each step is visible in the CI environment. At the end of every test, whether the integration passes or fails, you'll find screenshots of each step the agent took during the session, along with an
.mp4video recording that captures the entire session. By reviewing these recordings and screenshots, you can observe how the agent progressed through each stage, how long the task took, and what decisions were made based on the system prompt. You can use these examples as a reference when crafting your own system prompts or instructions for the MCP server in your own environment.
[!WARNING] Artifacts attached to these runs may have expired due to GitHub's artifact retention policy. Persistent copies are prepared via the Persist Artifacts workflow and can always be accessed by run ID from the
artifacts/directory on the press-kit branch.
Architecture
graph TB
subgraph MCP["MCP Client (Claude)"]
AI["Claude"]
end
subgraph Proxy["claude-kvm · MCP Proxy (stdio)"]
direction TB
Server["MCP Server<br/><code>index.js</code>"]
Tools["Tool Definitions<br/><code>tools/index.js</code>"]
Server --> Tools
end
subgraph Daemon["claude-kvm-daemon · Native VNC Client (stdin/stdout)"]
direction TB
CMD["Command Handler<br/><i>PC Dispatch</i>"]
Scale["Display Scaling<br/><i>Scaled ↔ Native</i>"]
subgraph Screen["Screen"]
Capture["Frame Capture<br/><i>PNG · Crop · Diff</i>"]
OCR["OCR Detection<br/><i>Apple Vision</i>"]
end
subgraph InputGroup["Input"]
Mouse["Mouse<br/><i>Click · Drag · Move · Scroll</i>"]
KB["Keyboard<br/><i>Tap · Combo · Type · Paste</i>"]
end
VNC["VNC Bridge<br/><i>LibVNCClient 0.9.15</i>"]
CMD --> Scale
Scale --> Capture
Scale --> Mouse
Scale --> KB
Capture -.->|"framebuffer"| VNC
Mouse -->|"pointer events"| VNC
KB -->|"key events"| VNC
end
subgraph Target["Target Machine"]
VNC_Server["VNC Server<br/><i>:5900</i>"]
Desktop["Desktop Environment"]
VNC_Server --> Desktop
end
AI <-->|"stdio<br/>JSON-RPC"| Server
Server <-->|"stdin/stdout<br/>PC (NDJSON)"| CMD
VNC <-->|"RFB Protocol<br/>TCP :5900"| VNC_Server
classDef proxy fill:#1a1a2e,stroke:#16213e,color:#e5e5e5
classDef daemon fill:#0f3460,stroke:#533483,color:#e5e5e5
classDef target fill:#1a1a2e,stroke:#e94560,color:#e5e5e5
class Server,Tools proxy
class CMD,Scale,VNC,Capture,Mouse,KB daemon
class VNC_Server,Desktop target
Layers
| Layer | Language | Role | Communication |
|---|---|---|---|
| MCP Proxy | JavaScript (Node.js) | Communicates with Claude over MCP protocol, manages daemon lifecycle | stdio JSON-RPC |
| VNC Daemon | Swift/C (Apple Silicon) | VNC connection, screen capture, mouse/keyboard input injection | stdin/stdout PC (NDJSON) |
PC (Procedure Call) Protocol
Communication between the proxy and daemon uses the PC protocol over NDJSON:
Request: {"method":"<name>","params":{...},"id":<int|string>}
Response: {"result":{...},"id":<int|string>}
Error: {"error":{"code":<int>,"message":"..."},"id":<int|string>}
Notification: {"method":"<name>","params":{...}}
Coordinate Scaling
The VNC server's native resolution is scaled down to fit within --max-dimension (default: 1280px). Claude works more consistently with scaled coordinates — the daemon handles the conversion in the background:
Native: 4220 x 2568 (VNC server framebuffer)
Scaled: 1280 x 779 (what Claude sees and targets)
mouse_click(640, 400) → VNC receives (2110, 1284)
Screen Strategy
Claude minimizes token cost with a progressive verification approach:
diff_check → changeDetected: true/false ~5ms (text only, no image)
detect_elements → OCR text + bounding boxes ~50ms (text only, no image)
cursor_crop → crop around cursor ~50ms (small image)
screenshot → full screen capture ~200ms (full image)
detect_elements uses Apple Vision framework for on-device OCR. Returns text content with bounding box coordinates in scaled space — enables precise click targeting without consuming vision tokens.
Installation
Requirements
- macOS (Apple Silicon / aarch64)
- Node.js (LTS)
Daemon
brew tap ARAS-Workspace/tap
brew install claude-kvm-daemon
[!NOTE]
claude-kvm-daemonis compiled and code-signed via CI (GitHub Actions). The build output is packaged in two formats: a.tar.gzarchive for Homebrew distribution and a.dmgdisk image for notarization. The DMG is submitted to Apple servers for notarization within the same workflow — the process can be tracked from CI logs. The notarized DMG is available as a CI Artifact; the archived.tar.gzis also published as a release on the repository. Homebrew installation tracks this release.
MCP Configuration
Create a .mcp.json file in your project directory:
{
"mcpServers": {
"claude-kvm": {
"command": "npx",
"args": ["-y", "claude-kvm"],
"env": {
"VNC_HOST": "192.168.1.100",
"VNC_PORT": "5900",
"VNC_USERNAME": "user",
"VNC_PASSWORD": "pass",
"CLAUDE_KVM_DAEMON_PATH": "/opt/homebrew/bin/claude-kvm-daemon",
"CLAUDE_KVM_DAEMON_PARAMETERS": "-v"
}
}
}
}
[!NOTE] The tool is end-to-end tested via CI — Claude executes tasks over VNC while an independent vision model observes and verifies the results. See the Integration Test for live workflow runs, system prompts, and demo recordings.
Configuration
MCP Proxy (ENV)
| Parameter | Default | Description |
|---|---|---|
VNC_HOST | 127.0.0.1 | VNC server address |
VNC_PORT | 5900 | VNC port number |
VNC_USERNAME | Username (required for ARD) | |
VNC_PASSWORD | Password | |
CLAUDE_KVM_DAEMON_PATH | claude-kvm-daemon | Daemon binary path (not needed if already in PATH) |
CLAUDE_KVM_DAEMON_PARAMETERS | Additional CLI arguments for the daemon |
Daemon Parameters (CLI)
Additional arguments passed to the daemon via CLAUDE_KVM_DAEMON_PARAMETERS:
"CLAUDE_KVM_DAEMON_PARAMETERS": "--max-dimension 800 -v"
| Parameter | Default | Description |
|---|---|---|
--max-dimension | 1280 | Maximum display scaling dimension (px) |
--connect-timeout | VNC connection timeout (seconds) | |
--bits-per-sample | Bits per pixel sample | |
--no-reconnect | Disable automatic reconnection | |
-v, --verbose | Verbose logging (stderr) |
Runtime Configuration (PC)
All timing and display parameters are configurable at runtime via the configure method. Use get_timing to inspect current values.
Set timing:
{"method":"configure","params":{"click_hold_ms":80,"key_hold_ms":50}}
{"result":{"detail":"OK — changed: click_hold_ms, key_hold_ms"}}
Change display scaling:
{"method":"configure","params":{"max_dimension":960}}
{"result":{"detail":"OK — changed: max_dimension","scaledWidth":960,"scaledHeight":584}}
Reset to defaults:
{"method":"configure","params":{"reset":true}}
{"result":{"detail":"OK — reset to defaults","timing":{"click_hold_ms":50,"combo_mod_ms":10,"cursor_crop_radius":150,"double_click_gap_ms":50,"drag_min_steps":10,"drag_pixels_per_step":20,"drag_position_ms":30,"drag_press_ms":50,"drag_settle_ms":30,"drag_step_ms":5,"hover_settle_ms":400,"key_hold_ms":30,"max_dimension":1280,"paste_settle_ms":30,"scroll_press_ms":10,"scroll_tick_ms":20,"type_inter_key_ms":20,"type_key_ms":20,"type_shift_ms":10},"scaledWidth":1280,"scaledHeight":779}}
Get current values:
{"method":"get_timing"}
{"result":{"timing":{"click_hold_ms":80,"combo_mod_ms":10,"cursor_crop_radius":150,"double_click_gap_ms":50,"drag_min_steps":10,"drag_pixels_per_step":20,"drag_position_ms":30,"drag_press_ms":50,"drag_settle_ms":30,"drag_step_ms":5,"hover_settle_ms":400,"key_hold_ms":50,"max_dimension":1280,"paste_settle_ms":30,"scroll_press_ms":10,"scroll_tick_ms":20,"type_inter_key_ms":20,"type_key_ms":20,"type_shift_ms":10},"scaledWidth":1280,"scaledHeight":779}}
| Parameter | Default | Description |
|---|---|---|
max_dimension | 1280 | Max screenshot dimension |
cursor_crop_radius | 150 | Cursor crop radius (px) |
click_hold_ms | 50 | Click hold duration |
double_click_gap_ms | 50 | Double-click gap delay |
hover_settle_ms | 400 | Hover settle wait |
drag_position_ms | 30 | Pre-drag position wait |
drag_press_ms | 50 | Drag press hold threshold |
drag_step_ms | 5 | Between interpolation pts |
drag_settle_ms | 30 | Settle before release |
drag_pixels_per_step | 20 | Point density per pixel |
drag_min_steps | 10 | Min interpolation steps |
scroll_press_ms | 10 | Scroll press-release gap |
scroll_tick_ms | 20 | Inter-tick delay |
key_hold_ms | 30 | Key hold duration |
combo_mod_ms | 10 | Modifier settle delay |
type_key_ms | 20 | Key hold during typing |
type_inter_key_ms | 20 | Inter-character delay |
type_shift_ms | 10 | Shift key settle |
paste_settle_ms | 30 | Post-clipboard write wait |
Tools
All operations are performed through a single vnc_command tool:
Screen
| Action | Parameters | Description |
|---|---|---|
screenshot | Full screen PNG capture | |
cursor_crop | Crop around cursor with crosshair overlay | |
diff_check | Detect screen changes against baseline | |
set_baseline | Save current screen as diff reference |
Mouse
| Action | Parameters | Description |
|---|---|---|
mouse_click | x, y, button? | Click (left|right|middle) |
mouse_double_click | x, y | Double click |
mouse_move | x, y | Move cursor |
hover | x, y | Move + settle wait |
nudge | dx, dy | Relative cursor movement |
mouse_drag | x, y, toX, toY | Drag from start to end |
scroll | x, y, direction, amount? | Scroll (up|down|left|right) |
Keyboard
| Action | Parameters | Description |
|---|---|---|
key_tap | key | Single key press (enter|escape|tab|space|...) |
key_combo | key or keys | Modifier combo ("cmd+c" or ["cmd","shift","3"]) |
key_type | text | Type text character by character |
paste | text | Paste text via clipboard |
Detection
| Action | Parameters | Description |
|---|---|---|
detect_elements | OCR text detection with bounding boxes (Apple Vision) |
Returns text elements with bounding box coordinates in scaled space:
{"method":"detect_elements"}
{"result":{"detail":"13 elements","elements":[{"confidence":1,"h":9,"text":"Finder","w":32,"x":37,"y":6},{"confidence":1,"h":9,"text":"File","w":15,"x":84,"y":6},{"confidence":1,"h":9,"text":"Edit","w":19,"x":112,"y":6},{"confidence":1,"h":9,"text":"View","w":22,"x":143,"y":6},{"confidence":1,"h":11,"text":"Go","w":15,"x":179,"y":6},{"confidence":1,"h":9,"text":"Window","w":35,"x":207,"y":6},{"confidence":1,"h":11,"text":"Help","w":22,"x":255,"y":6},{"confidence":1,"h":11,"text":"8•","w":26,"x":1161,"y":6},{"confidence":1,"h":9,"text":"Fri Feb 20 22:19","w":80,"x":1189,"y":6},{"confidence":1,"h":9,"text":"Assets","w":32,"x":1202,"y":97},{"confidence":1,"h":9,"text":"Passwords.kdbx","w":74,"x":1181,"y":168},{"confidence":1,"h":93,"text":"PHANTOM","w":633,"x":322,"y":477},{"confidence":1,"h":32,"text":"YOUR SERVER, YOUR NETWORK, YOUR PRIVACY","w":629,"x":325,"y":568}],"scaledHeight":717,"scaledWidth":1280}}
Configuration
| Action | Parameters | Description |
|---|---|---|
configure | {<params>} | Set timing/display params at runtime |
configure | {reset: true} | Reset all params to defaults |
get_timing | Get current timing + display params |
Control
| Action | Parameters | Description |
|---|---|---|
wait | ms? | Wait (default 500ms) |
health | Connection status + display info | |
shutdown | Graceful daemon shutdown |
Authentication
Supported VNC authentication methods:
- VNC Auth — password-based challenge-response (DES)
- ARD — Apple Remote Desktop (Diffie-Hellman + AES-128-ECB)
macOS is auto-detected via the ARD auth type 30 credential request. When detected, Meta keys are remapped to Super (Command key compatibility).
[!NOTE] Running on a bare-metal Mac? See the Mac M1 Preparation Tricks for VNC hardening, SSH tunneling, and session stability tips.
"Claude" is a trademark of Anthropic, PBC. This project is not affiliated with or endorsed by Anthropic.
Copyright (c) 2026 Riza Emre ARAS — MIT License
常见问题
Claude KVM 是什么?
用于通过 VNC 控制远程桌面的 MCP server,基于原生 Swift daemon,并集成 Apple Vision OCR。
相关 Skills
Claude接口
by anthropics
面向接入 Claude API、Anthropic SDK 或 Agent SDK 的开发场景,自动识别项目语言并给出对应示例与默认配置,快速搭建 LLM 应用。
✎ 想把Claude能力接进应用或智能体,用claude-api上手快、兼容Anthropic与Agent SDK,集成路径清晰又省心
提示工程专家
by alirezarezvani
覆盖Prompt优化、Few-shot设计、结构化输出、RAG评测与Agent工作流编排,适合分析token成本、评估LLM输出质量,并搭建可落地的AI智能体系统。
✎ 把提示优化、LLM评测到RAG与智能体设计串成一套方法,适合想系统提升AI开发效率的人。
智能体流程设计
by alirezarezvani
面向生产级多 Agent 编排,梳理顺序、并行、分层、事件驱动、共识五种工作流设计,覆盖 handoff、状态管理、容错重试、上下文预算与成本优化,适合搭建复杂 AI 协作系统。
✎ 帮你把多智能体流程设计、编排和自动化统一起来,复杂工作流也能更稳地落地,适合追求强控制力的团队。
相关 MCP Server
顺序思维
编辑精选by Anthropic
Sequential Thinking 是让 AI 通过动态思维链解决复杂问题的参考服务器。
✎ 这个服务器展示了如何让 Claude 像人类一样逐步推理,适合开发者学习 MCP 的思维链实现。但注意它只是个参考示例,别指望直接用在生产环境里。
知识图谱记忆
编辑精选by Anthropic
Memory 是一个基于本地知识图谱的持久化记忆系统,让 AI 记住长期上下文。
✎ 帮 AI 和智能体补上“记不住”的短板,用本地知识图谱沉淀长期上下文,连续对话更聪明,数据也更可控。
PraisonAI
编辑精选by mervinpraison
PraisonAI 是一个支持自反思和多 LLM 的低代码 AI 智能体框架。
✎ 如果你需要快速搭建一个能 24/7 运行的 AI 智能体团队来处理复杂任务(比如自动研究或代码生成),PraisonAI 的低代码设计和多平台集成(如 Telegram)让它上手极快。但作为非官方项目,它的生态成熟度可能不如 LangChain 等主流框架,适合愿意尝鲜的开发者。

