Skip to main content
Glama
Sign In
@qirabot/mcp

@qirabot/mcp

by qirabot
OverviewSchemaSourceRelated ServersDiscussionsAdmin
TypeScript
Remote

@qirabot/mcp

MCP (Model Context Protocol) server for Qirabot — AI-powered device automation for Claude, Cursor, and other MCP clients.

Control mobile and desktop devices with natural language through your AI assistant. The server exposes tools for running automation tasks, taking screenshots, and managing devices.

Installation

npm install -g @qirabot/mcp

Configuration

  1. Sign up at qirabot.com and get your API key from the dashboard.

  2. Set it as an environment variable:

export QIRA_API_KEY="qk_your_api_key"

Setup

Claude Desktop

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "qira": {
      "command": "qira-mcp-server",
      "env": {
        "QIRA_API_KEY": "qk_your_api_key"
      }
    }
  }
}

Claude Code

claude mcp add qira -- qira-mcp-server --api-key $QIRA_API_KEY

Cursor

Add to your Cursor MCP settings (.cursor/mcp.json):

{
  "mcpServers": {
    "qira": {
      "command": "qira-mcp-server",
      "env": {
        "QIRA_API_KEY": "qk_your_api_key"
      }
    }
  }
}

Remote HTTP Mode

For shared or remote deployments, run the server in HTTP mode:

qira-mcp-server --transport http --port 3100

Clients connect with API key in the header:

{
  "mcpServers": {
    "qira": {
      "url": "https://your-server/mcp",
      "headers": {
        "x-qira-api-key": "qk_your_api_key"
      }
    }
  }
}

Tools

Tool

Description

list_devices

List all active devices with their IDs, OS, type, and resolution

list_model_aliases

List available AI model aliases (e.g. 'balanced', 'high-quality')

run_task

Run an automation task on a device with a natural language instruction

get_task

Get the status and step details of a running or completed task

take_screenshot

Capture a live screenshot of a device's screen

cancel_task

Cancel a pending or running task

Usage Examples

Once configured, you can ask your AI assistant:

  • "List my connected devices"

  • "Take a screenshot of my phone"

  • "Open Chrome on my device and search for 'weather today'"

  • "Go to wikipedia.org and extract the featured article title"

  • "Open the Settings app and enable dark mode"

CLI Options

qira-mcp-server [options]

Options:
  --server, -s      Server URL (default: https://app.qirabot.com, or QIRA_SERVER_URL env)
  --api-key, -k     API key (required for stdio; or QIRA_API_KEY env)
  --transport, -t   Transport mode: "stdio" (default) or "http"
  --port, -p        HTTP server port (default: 3100)
  --help, -h        Show help

Environment Variables

Variable

Description

QIRA_SERVER_URL

Server URL (default: https://app.qirabot.com)

QIRA_API_KEY

API key (alternative to --api-key)

License

MIT

Available Tools

6 tools
cancel_taskA

Cancel a task that is currently in 'pending' or 'running' status. Has no effect on tasks that have already completed (succeeded, failed, timeout).

Inputs: task_id (required).

get_taskA

Get the status and step details of a task started by run_task.

Returns current status and all completed steps with action type, AI decision, params, and output. Possible status values: pending, running, succeeded, failed, cancelled, timeout, partially_succeeded. If status is 'pending' or 'running', poll again (recommended interval: 2-3 seconds).

Screenshots are not included in the response — use take_screenshot after completion to verify the result.

Inputs: task_id (required).

list_devicesA

List all active devices connected to the server. Returns device ID, name, OS, type, screen resolution and host info. Call this first to get the device_id required by take_screenshot, run_task, and other device tools.

Inputs: no named parameters.

list_model_aliasesA

List available model aliases (e.g. 'balanced', 'high-quality'). Use the alias name as the model_alias parameter in run_task.

Inputs: no named parameters.

run_taskA

Perform UI automation on a device. The AI agent automatically handles app launching, screen interactions (tapping, swiping, typing, navigating), and multi-step workflows.

Use this tool for:

  • Opening apps and performing tasks: 'open Chrome and search for weather'

  • In-app interactions: tapping, swiping, typing, navigating

  • Content engagement: liking, commenting, sharing, following, subscribing

  • Information gathering: reading screen content, searching, extracting data

  • Web browsing: visiting URLs, filling forms, clicking links

  • Messaging: sending messages, replying, forwarding

Do NOT use this tool to only take a screenshot — use take_screenshot instead.

Provide only the end goal in instruction (e.g., 'open WeChat and send hello to Zhang San'), not low-level steps. Returns immediately with a task ID. Use get_task to poll for status and results. After completion, call take_screenshot to verify the result.

Inputs: device_id (required), instruction (required), model_alias, max_steps.

take_screenshotA

Capture a live screenshot of the device's current screen. Returns the image directly (synchronous, up to 30s).

Use this tool when:

  • You need to see what is currently displayed on the device screen.

  • You want to verify the result after a run_task completes.

  • You need to inspect the UI before deciding what action to take.

This returns the real-time screen.

Inputs: device_id (required).

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Deploy Server
    One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

    TDQS

    A4.4/5.0

    Scored across 6 tools

    Disambiguation5/5

    Each tool has a unique and clearly defined purpose: starting tasks, monitoring, canceling, capturing screenshots, listing devices, and listing model aliases. There is no overlap or ambiguity.

    Naming Consistency5/5

    All tool names follow a consistent verb_noun pattern using snake_case (e.g., run_task, get_task, list_devices). No mixed conventions or vague verbs.

    Tool Count5/5

    With 6 tools, the set is well-scoped for UI automation on mobile devices. Each tool serves a necessary function without being excessive or insufficient.

    Completeness4/5

    The core workflow (run, monitor, cancel, screenshot) is covered, along with device and model configuration. A minor gap is the lack of a tool to list all active tasks, but this does not severely hinder functionality.

    Maintenance

    ActivityNo data
    ResponsivenessSyncing

    Resources

    Related MCP Connectors