screen-mcp
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@screen-mcptake a screenshot of my screen"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
screen-mcp
MCP local (stdio) que dá ao Claude controle da sua tela no Windows: ver, achar por imagem, clicar/digitar, gravar e repetir macros, e agendar. Sem porta exposta.
Veja SPEC.md pro design completo.
Instalar
python -m venv .venv
.venv\Scripts\python.exe -m pip install -r requirements.txtRegistre no seu ~/.claude.json:
{
"mcpServers": {
"screen-mcp": {
"command": "<caminho>\\.venv\\Scripts\\python.exe",
"args": ["<caminho>\\screenmcp\\cli.py", "serve"]
}
}
}Reinicie o Claude Code pra ele carregar o server. Depois é só pedir em linguagem natural ("tira um print", "acha o botão X e clica", "grava uma macro", etc.).
Related MCP server: device-controller-mcp
Ferramentas
Ver/achar:
list_monitors,screenshot,screen_info,find_image,wait_for_imageAgir:
move,click,drag,type_text,press_keys,scrollMacros:
list_macros,record_macro,replay_macro,delete_macroAgendar:
schedule_macro,list_schedules,unschedule
Os 4 modos
Ao vivo — eu uso as ferramentas vendo a tela e decidindo.
Macro gravada —
record_macro("nome")abre uma janela; você faz as ações; F9 salva. Depoisreplay_macro("nome", times=N).Loop com condição —
wait_for_image(...)+ ação, repetido (ex: refresh até um botão aparecer).Agendado —
schedule_macro("nome", "DAILY:09:00")cria tarefa no Agendador do Windows (MINUTE:N|HOURLY|DAILY:HH:MM|ONLOGON); roda sem o Claude presente.
Achar por imagem (o jeito robusto)
Salve um recorte (.png) do elemento em templates/ e use find_image("botao.png") → devolve x,y globais do centro pra click. Resistente a janela que se move, melhor que chutar coordenada.
Segurança
Mouse no canto superior-esquerdo aborta tudo na hora (failsafe).
F12 aborta um replay em andamento.
Teto de loops + pausa entre ações. Log em
logs/actions.jsonl.
Rodar standalone (sem Claude)
.venv\Scripts\python.exe screenmcp\cli.py list
.venv\Scripts\python.exe screenmcp\cli.py record minha_macro
.venv\Scripts\python.exe screenmcp\cli.py replay minha_macro --times 5Reinstalar deps
.venv\Scripts\python.exe -m pip install -r requirements.txtCoordenadas / multi-monitor
Tudo em coords globais do virtual desktop (monitor primário = origem; monitores à esquerda têm x negativo). Processo é per-monitor DPI aware. Em setups com DPI mista entre monitores, prefira find_image a chutar coordenada.
Licença
MIT
Related MCP Connectors
Hosted MCP server connecting claude.ai, ChatGPT and other AI apps to your own computer
Remote MCP server for supportsheep: run AI interviews and manage support content for your blog.
Augments MCP Server - A comprehensive framework documentation provider for Claude Code
MCP server for progressive tool usage at any scale (see https://klavis.ai)
Related MCP Servers
- FlicenseAqualityDmaintenanceAn MCP server that provides Claude with comprehensive desktop automation capabilities including browser control, window management, and native mouse/keyboard input on Windows. It enables users to capture screenshots, launch applications, and interact with the system clipboard through natural language.20-
- AlicenseNot gradedqualityBmaintenanceAn MCP server that lets Claude Desktop and Claude Code control your PC — take screenshots, click, type, manage windows, and more.MIT
- AlicenseNot gradedqualityCmaintenanceA local, dependency-free MCP server that gives AI agents controlled access to the active Windows desktop, enabling automated interaction with applications through screenshots, clicks, typing, and window management.87 npmMIT
- AlicenseNot gradedqualityBmaintenanceA Windows desktop control MCP server that gives Claude eyes and hands—screen capture with coordinate-grid overlays, pixel-perfect DPI-correct mouse/keyboard input via a Rust backend, and direct PowerShell command execution, enabling full local desktop automation through natural language.28 npm1MIT