Overview

Cursormcp.json / Cursor settings, tools appear in the agent panel.Claude CodeCLI agent, claude mcp add or project .mcp.json.Claude DesktopDesktop app config, restart after wiring Screen Beaver.WindsurfCascade Plugins UI or mcp_config.json.TraeByteDance Trae, project .trae/mcp.json.AntigravityGoogle Antigravity, Gemini MCP config.CodexOpenAI Codex CLI, config.toml or codex mcp add.GitHub CopilotVS Code Copilot Chat, .vscode/mcp.json.VS CodeNative MCP in VS Code, same as Copilot wiring.ClineCline marketplace / cline_mcp_settings.json.ContinueContinue.dev, MCP in config.yaml / IDE settings.Roo CodeRoo Code extension, MCP server list.KiroAWS Kiro, MCP via IDE settings.OpenCodeOpenCode agent, MCP server entry.WarpWarp terminal agents, MCP launch config.ZedZed editor, context server / MCP settings.Amazon QAmazon Q Developer, ~/.aws/amazonq/mcp.json.JetBrains AIJetBrains AI Assistant, Settings → MCP.AugmentAugment Code, MCP integrations panel.Gemini CLIGemini CLI / Google tooling, MCP config.npmGlobal CLI install

Screen Beaver MCP lets coding agents drive the product-demo studio while you watch. Agents open targets, click through flows, plant zooms and captions, generate voiceover, and prepare export, so you can make tutorials without touching your screen.

What you get

• Hands-off capture across tools · Facts you ask in any vibe coding agent become real clicks and typing inside Screen Beaver, recorded and auto-polished in the same pipeline.
• Same inspectors agents and humans share · Background blur, zooms, cursor polish, captions, effects, and export stay in one studio UI. MCP calls land live while you watch.
• Local export, always · Encoding stays on your machine as H.264 MP4. Nothing is uploaded for export.
• Optional desktop unlocks · The Windows app adds system-wide cursor tracking, picker-free native capture, USB Android recording, and full MCP control.

Why a small tool surface is intentional

Agents get high-leverage flows like record_and_build and live_build_demo instead of hundreds of micro-APIs. That keeps demos reliable: capture → auto polish → you hit Export.

How it fits together

Install the Windows desktop app first, MCP only works after Screen Beaver is installed and running. Then install the MCP CLI, point your vibe coding agent at it, and ask the agent to film a feature walkthrough. Screen Beaver records the session, runs the auto pipeline (zooms, SFX, captions), and opens the timeline for review. Manual studio controls remain available whenever you want to drive it yourself.

Connect

Order matters: Get from Microsoft Store → install the app → then add Screen Beaver to your MCP client. Full steps in Setup and Usage.

123456789npm i -g @screenbeaver/mcp{  "mcpServers": {    "screenbeaver": {      "command": "screenbeaver-mcp"    }  }}

Tools

• Studio presence · ping, status, capabilities, focus, window layout.
• Capture · list targets, open URL/path, start/stop native capture.
• Demo input · move, click, type, scroll, wait, run scripted demos during an active capture.
• Edit · seek, captions, zooms, effects, SFX, camera moves, regenerate zooms, apply look presets.
• Voice & music · translate captions, Edge neural TTS, Creative Commons beds.
• Ship · ready_for_export, export_mp4.

Resources

Agents can read the in-product guide via screen-weaver://guide / get_agent_guide. Marketing overview: MCP agents page.

Prompts

Prefer short, outcome-shaped asks: “Record a 60s onboarding demo of the pricing page, add captions, and prepare export.” The agent should use record_and_build or live_build_demo, then let you confirm Export.