Agent setup¶

Every MCP-capable agent takes the same server block. Only the file it goes in changes.

{
  "mcpServers": {
    "screensight": {
      "command": "screensight-mcp"
    }
  }
}

Agent	Config file
Claude Code	`~/.claude.json` or project `.mcp.json`
Cursor	`~/.cursor/mcp.json`
Windsurf	`~/.codeium/windsurf/mcp_config.json`
Cline	VS Code settings, under `cline.mcpServers`
Codex CLI	`~/.codex/config.json`
Aider / no MCP	No config — call the CLI directly

Ready-made snippets for each agent also ship in the repo under examples/.

Claude Code¶

Project-level (recommended)GlobalLegacy skill adapter

Create .mcp.json in your project root:

{
  "mcpServers": {
    "screensight": {
      "command": "screensight-mcp"
    }
  }
}

Add to ~/.claude.json:

{
  "mcpServers": {
    "screensight": {
      "command": "screensight-mcp"
    }
  }
}

bash install.sh --with-claude-skill

Copies the legacy skill to ~/.claude/skills/SKILL.md. Optional — the MCP server is the supported path.

Cursor¶

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "screensight": {
      "command": "screensight-mcp"
    }
  }
}

Restart Cursor after saving.

Windsurf¶

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "screensight": {
      "command": "screensight-mcp"
    }
  }
}

Restart Windsurf after saving.

Cline¶

Add to VS Code settings.json:

{
  "cline.mcpServers": {
    "screensight": {
      "command": "screensight-mcp"
    }
  }
}

Restart VS Code after saving.

Codex CLI¶

Add the same mcpServers block to ~/.codex/config.json.

Aider and other non-MCP agents¶

No configuration is needed — the agent shells out to the CLI. Put this in your agent instructions or system prompt:

When I ask you to look at my screen, run:
  screensight capture

The output will contain a "path" field with the screenshot location.
Read that file to see what's on my screen.

The switch still applies

Wiring up an agent does not enable capture. The master switch is off until you run screensight on — or until the agent calls screen_enable and you allow it.

Verify the connection¶

Ask the agent "what's on my screen?". It should call screen_capture and describe the frame. If the tool doesn't appear at all:

which screensight-mcp    # Linux / macOS
where screensight-mcp    # Windows

If the binary isn't found, the agent can't spawn it — reinstall with pip install -e . or use an absolute path in the command field. More in Troubleshooting.