ComputerUse CLI

The ComputerUse CLI is a powerful command-line tool for managing the ComputerUse project, including version management, releases, Azure VM deployment, and MCP server interaction.

Features

📦 Version Management: Bump and sync versions across all packages
🏷️ Release Automation: Tag and release with a single command
☁️ Azure VM Deployment: One-liner to deploy Windows VMs with MCP server
🤖 MCP Client: Chat with MCP servers over HTTP or stdio
🔄 Workflow Execution: Run automation workflows from YAML/JSON files
🔧 Tool Execution: Execute individual MCP tools directly
🔒 Secure by Default: Auto-generated passwords, configurable security rules

Installation

Windows (Recommended - npm wrapper):

bash

# Run directly without installation
npx @mediar-ai/cli --help
bunx @mediar-ai/cli --help

# Or install globally
npm install -g @mediar-ai/cli

macOS / Linux (Compile from Source):

⚠️ The npm package @mediar-ai/cli only includes Windows binaries. Other platforms must compile from source.

bash

# From the workspace root
cargo build --release --bin computeruse

# Install globally (optional)
cargo install --path crates/computeruse-cli

Quick Start

Version Management

bash

# Bump version
computeruse patch      # x.y.Z+1
computeruse minor      # x.Y+1.0
computeruse major      # X+1.0.0

# Sync all package versions
computeruse sync

# Show current status
computeruse status

# Tag and push
computeruse tag

# Full release (bump + tag + push)
computeruse release        # patch release
computeruse release minor  # minor release

MCP Workflow Execution

Execute automation workflows from YAML or JSON files:

bash

# Execute a workflow file
computeruse mcp run workflow.yml

# Execute with verbose logging
computeruse mcp run workflow.yml --verbose

# Dry run (validate without executing)
computeruse mcp run workflow.yml --dry-run

# Use specific MCP server command
computeruse mcp run workflow.yml --command "npx -y computeruse-mcp-agent@latest"

# Use HTTP MCP server
computeruse mcp run workflow.yml --url http://localhost:3000/mcp

# Pass input values to workflow (available as env variables in scripts)
computeruse mcp run workflow.yml --inputs '{"username":"john","api_key":"abc123"}'

# Combine inputs with other options
computeruse mcp run workflow.yml --inputs '{"count":5}' --verbose

Passing Input Values to Workflows

The --inputs parameter allows you to pass initial values to your workflow that can be accessed by JavaScript/Python scripts:

bash

# Pass inputs as JSON
computeruse mcp run workflow.yml --inputs '{"user":"alice","count":42,"enabled":true}'

These inputs are accessible in your workflow scripts:

In JavaScript: via env.inputs object or directly as env.username, env.count, etc.
In Python: via env dictionary
Inputs override any default values defined in the workflow file

Example workflow file (workflow.yml):

yaml

tool_name: execute_sequence
arguments:
  steps:
    - tool_name: navigate_browser
      arguments:
        url: "https://example.com"
    - tool_name: click_element
      arguments:
        selector: "role:Button && name:Submit"
    - tool_name: get_applications_and_windows_list
      id: get_apps
    - tool_name: run_command
      engine: javascript
      id: extract_pid
      run: |
        const apps = get_apps_result[0]?.applications || [];
        const focused = apps.find(app => app.is_focused);
        return { pid: focused?.pid || 0 };
    - tool_name: get_window_tree
      arguments:
        pid: "{{extract_pid.pid}}"
      id: capture_result
  output_parser:
    ui_tree_source_step_id: capture_result
    javascript_code: |
      // Extract all checkbox names
      const results = [];
      
      function findElementsRecursively(element) {
          if (element.attributes && element.attributes.role === 'CheckBox') {
              const item = {
                  name: element.attributes.name || ''
              };
              results.push(item);
          }
          
          if (element.children) {
              for (const child of element.children) {
                  findElementsRecursively(child);
              }
          }
      }
      
      findElementsRecursively(tree);
      return results;

JavaScript execution in workflows:

yaml

tool_name: execute_sequence
arguments:
  steps:
    - tool_name: run_command
      arguments:
        engine: "javascript"
        run: |
          // Access inputs passed from CLI via --inputs parameter
          console.log(`Processing for user: ${env.username}`);
          console.log(`Count value: ${env.count}`);

          // Or access the entire inputs object
          const allInputs = env.inputs;
          console.log(`All inputs:`, JSON.stringify(allInputs));

          // Use inputs in your logic
          for (let i = 0; i < env.count; i++) {
            console.log(`Processing item ${i + 1} for ${env.username}`);
          }

          return {
            processed_by: env.username,
            items_processed: env.count
          };

Original example:

yaml

tool_name: execute_sequence
arguments:
  steps:
    - tool_name: run_command
      arguments:
        engine: "node"
        script: |
          // Access desktop automation APIs
          const elements = await desktop.locator('role:button').all();
          log(`Found ${elements.length} buttons`);
          
          // Interact with UI elements
          for (const element of elements) {
            const name = await element.name();
            if (name.includes('Submit')) {
              await element.click();
              break;
            }
          }
          
          return {
            buttons_found: elements.length,
            action: 'clicked_submit'
          };

MCP Tool Execution

Execute individual MCP tools directly:

bash

# Execute a single tool
computeruse mcp exec get_applications

# Execute with arguments
computeruse mcp exec click_element '{"selector": "role:Button && name:OK"}'

# Use different MCP server
computeruse mcp exec --url http://localhost:3000/mcp validate_element '{"selector": "#button"}'

Interactive MCP Chat

Chat with MCP servers interactively:

bash

# Start chat session (uses local MCP server by default)
computeruse mcp chat

# Chat with remote MCP server
computeruse mcp chat --url https://your-server.com/mcp

# Chat with specific MCP server command
computeruse mcp chat --command "node my-mcp-server.js"

Control Remote Computer Through Chat

Run the MCP server on your remote machine
Open port or use ngrok
Connect via CLI:

bash

computeruse mcp chat --url https://xxx/mcp

Advanced Usage

MCP Server Connection Options

The CLI supports multiple ways to connect to MCP servers:

bash

# Local MCP server (default - uses @latest for compatibility)
computeruse mcp run workflow.yml

# Specific version
computeruse mcp run workflow.yml --command "npx -y [email protected]"

# HTTP server
computeruse mcp run workflow.yml --url http://localhost:3000/mcp

# Custom server command
computeruse mcp run workflow.yml --command "python my_mcp_server.py"

Workflow File Formats

The CLI supports both YAML and JSON workflow files:

Direct workflow (workflow.yml):

yaml

steps:
  - tool_name: navigate_browser
    arguments:
      url: "https://example.com"
stop_on_error: true

Tool call wrapper (workflow.json):

json

{
  "tool_name": "execute_sequence",
  "arguments": {
    "steps": [
      {
        "tool_name": "navigate_browser",
        "arguments": {
          "url": "https://example.com"
        }
      }
    ]
  }
}

Error Handling

bash

# Continue on errors
computeruse mcp run workflow.yml --continue-on-error

# Custom timeout
computeruse mcp run workflow.yml --timeout 30000

# Detailed error output
computeruse mcp run workflow.yml --verbose

Code Execution in Workflows

The CLI supports executing code within workflows using the run_command tool in engine mode, providing access to desktop automation APIs:

Available Engines:

nodejs - Full Node.js runtime with desktop APIs
quickjs - Lightweight JavaScript engine (default)

Desktop APIs Available:

javascript

// Element discovery
const elements = await desktop.locator('role:Button && name:Submit').all();
const element = await desktop.locator('#button-id').first();

// Element interaction
await element.click();
await element.type('Hello World');
await element.setToggled(true);

// Property access
const name = await element.name();
const bounds = await element.bounds();
const isEnabled = await element.enabled();

// Utilities
log('Debug message');  // Logging
await sleep(1000);     // Delay in milliseconds

Example Use Cases:

yaml

# Conditional logic based on UI state
- tool_name: run_command
  arguments:
    engine: "node"
    script: |
      const submitButton = await desktop.locator('role:Button && name:Submit').first();
      const isEnabled = await submitButton.enabled();
      
      if (isEnabled) {
        await submitButton.click();
        return { action: 'submitted' };
      } else {
        log('Submit button is disabled, checking form validation...');
        return { action: 'validation_needed' };
      }

# Bulk operations on multiple elements
- tool_name: run_command
  arguments:
    engine: "javascript"
    script: |
      const checkboxes = await desktop.locator('role:checkbox').all();
      let enabledCount = 0;
      
      for (const checkbox of checkboxes) {
        await checkbox.setToggled(true);
        enabledCount++;
        await sleep(50); // Small delay between operations
      }
      
      return { total_enabled: enabledCount };

# Dynamic element discovery and interaction
- tool_name: run_command
  arguments:
    engine: "javascript"
    script: |
      // Find all buttons containing specific text
      const buttons = await desktop.locator('role:button').all();
      const targets = [];
      
      for (const button of buttons) {
        const name = await button.name();
        if (name.toLowerCase().includes('download')) {
          targets.push(name);
          await button.click();
          await sleep(1000);
        }
      }
      
      return { downloaded_items: targets };

Configuration

Environment Variables

RUST_LOG: Set logging level (e.g., debug, info, warn, error)
MCP_SERVER_URL: Default MCP server URL
MCP_SERVER_COMMAND: Default MCP server command

Default Behavior

MCP Server: Uses npx -y computeruse-mcp-agent@latest by default
Logging: Info level by default, debug with --verbose
Error Handling: Stops on first error by default
Format: Auto-detects YAML/JSON from file extension

Troubleshooting

Version Mismatch Issues

If you encounter "missing field" errors, ensure you're using the latest MCP server:

bash

# Force latest version
computeruse mcp run workflow.yml --command "npx -y computeruse-mcp-agent@latest"

# Clear npm cache if needed
npm cache clean --force

Connection Issues

bash

# Test MCP server connectivity
computeruse mcp exec get_applications

# Use verbose logging for debugging
computeruse mcp run workflow.yml --verbose

# Test with dry run first
computeruse mcp run workflow.yml --dry-run

JavaScript Execution Issues

bash

# Test JavaScript execution capability via run_command (engine mode)
computeruse mcp exec run_command '{"engine": "javascript", "run": "return {test: true};"}'

# Use node engine for full APIs
computeruse mcp exec run_command '{"engine": "node", "run": "const elements = await desktop.locator(\"role:button\").all(); return {count: elements.length};"}'

# Run Python with computeruse.py
computeruse mcp exec run_command '{"engine": "python", "run": "return {\"ok\": True}"}'

# Debug JavaScript errors with verbose logging
computeruse mcp run workflow.yml --verbose

For more examples and advanced usage, see the ComputerUse MCP Agent documentation.