How to Manage and Automate Cloud Phones with GeeLark MCP
Share with
Summarize this article with your preferred AI
Meet GeeLark MCP Server. You can now connect GeeLark to AI agents like Codex, Claude, and Cursor. Tell the AI what you want to do, and it can help manage your cloud phones and browser profiles or set up automation tasks.
This guide explains what GeeLark MCP does and how to connect it to you AI agent. I’ll also share real examples of using AI to manage cloud phones and automate everyday tasks.
What is GeeLark MCP
GeeLark MCP lets AI agents like Codex, Claude, and Cursor use the GeeLark API to work with your cloud phones, browser profiles, and other features.
It uses Model Context Protocol (MCP), an open standard that connects AI tools to other services. Once you connect GeeLark and authorize it with an API token, you can tell your AI what to do in plain English. The AI then calls the GeeLark tools needed for the task.
Here’s what GeeLark MCP supports:
| Category | What you can do |
| Cloud phones | Find, create, edit, start, stop, duplicate, and delete cloud phones. Check device status and take screenshots. |
| Browser profiles | Create, manage, and launch browser profiles, take screenshots, and schedule browser automation tasks. |
| Proxies | Add, view, edit, delete, and test proxies. |
| Groups | Create, view, edit, and delete groups to organize devices by project. |
| Tags | Create, view, edit, and delete tags, and assign them to cloud phones. |
| Automation | Create and schedule automation tasks, check task history, cancel or rerun tasks, and use supported social media automation templates. |
| Apps and files | Install and remove apps, manage apps across multiple devices, upload files, and manage content in GeeLark Library. |
| Billing and plans | Check your balance, transactions, and plan details. Use supported options to renew or upgrade a plan and buy cloud phone time. |
Before you connect to GeeLark MCP
Before you start, you’ll need:
- A paid GeeLark account: Make sure your plan includes API access (Base or Pro).
- GeeLark API token: You can find it in GeeHub in the GeeLark app.
- An AI agent that supports MCP: For example, Codex, Claude, Cursor, or another AI agent.
In GeeHub, make sure API Connection shows Active. Then find GEELARK_TOKEN in the MCP settings. You’ll need that token when you set up the connection.
Keep it private. Don’t paste your token into the chat.
Connect to GeeLark MCP
Set up GeeLark MCP with AI
The easiest way is to open your AI agent and send it this prompt:
Help me connect GeeLark MCP Server to the AI agent I’m using. Assume I’m new to this and walk me through setup one step at a time.
Start by reading the official GeeLark documentation:
https://github.com/GeeLark/geelark-cli/#use-as-mcp-server
Here’s how I’d like you to help:
1. Go step by step. Don’t give me a long list of commands or config files upfront. Explain only the next step, wait until I finish it, and use simple language.
2. Do as much of the setup as you can. First, check my computer and choose a suitable installation method. If you can make changes on my computer, ask before installing software or changing settings. If you can’t, tell me exactly where to click and what to enter.
3. Help me enter the GeeLark token locally. In the GeeLark app, open GeeHub → API → MCP. Look for GEELARK_TOKEN. Show me how to enter it on my computer. Don’t ask me to paste my real token into this chat.
4. Follow the official documentation. Don’t guess at settings or use a third-party MCP server. If you can’t access the docs, ask me to paste the relevant section.
5. Test the connection. When setup is complete, check that GeeLark MCP works by trying to retrieve my cloud phone list.
Most importantly, don’t send me the entire setup process at once. Tell me only the first step for now.
Then follow your AI agent’s instructions to complete the setup.
Set up GeeLark MCP manually
If you’ve set up MCP servers before, you can add GeeLark MCP directly to your AI tool.
Claude Code/Cursor/Trae
Open your AI tool’s MCP configuration file. For example:
- Cursor: \~/.cursor/mcp.json
- Trae: \~/.trae/mcp/mcp.json
- Claude Code: You can add the server with the claude mcp add command.
For tools that use JSON configuration files, you can start with this example:
{
“mcpServers”: {
“geelark”: {
“command”: “npx”,
“args”: [“-y”, “geelark-cli@latest”, “mcp”],
“env”: {
“GEELARK_TOKEN”: “YOUR_GEELARK_TOKEN”
}
}
}
}
Replace YOUR_GEELARK_TOKEN with the token from GeeHub → MCP in the GeeLark app.
This setup runs GeeLark MCP with npx, so your computer needs a working Node.js environment.
GeeLark also supports two optional settings:
- GEELARK_BASE_URL: The GeeLark API address. The default is https://openapi.geelark.com.
- GEELARK_BROWSER_BASE_URL: The local browser API address. The default is http://localhost:40185.
If you don’t need custom addresses, you can leave these settings out.
If you already have geelark-cli installed, you can start the server with geelark-cli mcp instead of npx.
Reload your MCP client, then ask your AI agent to list your GeeLark cloud phones to make sure the connection works.
For more setup options, see the official GeeLark MCP documentation.
Codex
Open the Codex configuration file at \~/.codex/config.toml. Add the example for your operating system.
Windows:
[mcp_servers.geelark]
command = “cmd.exe”
args = [“/d”, “/c”, “npx”, “-y”, “geelark-cli@latest”, “mcp”]
[mcp_servers.geelark.env]
GEELARK_TOKEN = “YOUR_GEELARK_TOKEN”
macOS / Linux:
[mcp_servers.geelark]
command = “npx”
args = [“-y”, “geelark-cli@latest”, “mcp”]
[mcp_servers.geelark.env]
GEELARK_TOKEN = “YOUR_GEELARK_TOKEN”
Replace YOUR_GEELARK_TOKEN with the token from GeeHub → MCP.
Save the file and restart Codex. You can now try using GeeLark MCP.
Note: Your computer needs to run npx for this setup. If the connection fails, ask your AI agent to check your computer’s setup and the MCP configuration.
Test Your Connection
Once you’re set up, send this simple prompt to test the connection:
How many cloud phones do I have in GeeLark?
If the connection works, your AI will use GeeLark MCP to count your cloud phones.
Compare that number with the one shown in the GeeLark app. If they match, you’re ready to go.
Get your AI ready to use GeeLark MCP
Different AI agents and models may understand instructions and use tools in different ways. After you connect MCP, your AI can usually see the available tools. But it may not know right away how to combine them to finish a full task.
In my tests, things went more smoothly when I first asked the AI to learn what GeeLark MCP could do. That way, I didn’t have to remember API commands or exact tool names. I could describe the job in everyday language instead.
Start with this prompt:
I’ve connected GeeLark MCP to my AI agent.
Before we start, please check the GeeLark MCP tools available to you. Understand what each tool does, how to use it, and any important limitations.
I’ll describe what I want to do in everyday language, without knowing the exact tool names or API commands. Please figure out which GeeLark tools to use and how to combine them when a task requires multiple steps.
Only use tools that are actually available. Don’t guess what GeeLark MCP can do. Ask me questions only when important information is missing, and get my confirmation before making changes.
For now, don’t perform any actions. Just give me a short overview of what you can help me do, along with five practical examples.
Once that’s done, you can ask the AI to find certain cloud phones, organize devices by project, or schedule automation tasks. It can choose from the GeeLark MCP tools that are actually available.
GeeLark MCP use cases
GeeLark MCP is useful for more than asking AI to click buttons. It can help you connect device searches, information cleanup, and actions in one workflow. Here are seven practical examples. You can copy the prompts and replace the text in [brackets] with your own details.
1. Set up cloud phones
Say you need 10 cloud phones running Android 13–16. You want different phone brands and models, plus five US proxies from the ones you’ve already saved in GeeLark. Each phone should also have a GPS location that matches its assigned proxy.
Doing this by hand means checking the available Android versions, models, proxies, and locations, then setting up the phones. With GeeLark MCP, you can ask AI to check what’s available, prepare a plan, and create the phones after you approve it.
One detail matters here: a proxy check may show the country, state, and city of its IP address, but not exact GPS coordinates. The AI should look up reliable coordinates rather than make them up. Also, if you spread five proxies across 10 phones, some proxies will be shared. Review that plan before creating the phones.
Prompt:
Create [10] cloud phones with Android [13–16] and different available models. Pick [5] working US proxies from my saved list and assign them across the phones. Set each phone’s GPS to match its proxy IP’s detected location, using verified coordinates.
2. Analyze your cloud phones
When you manage dozens or hundreds of cloud phones, it’s easy to lose track of your device mix. How many run Android 14? How many use Android 16? Which brands and models do you use most?
Ask your AI to read your GeeLark device list, count phones by Android version, brand, and model, and put the results in a table. It’s a quick way to see your device setup without checking phones one by one.
Prompt:
In [YOUR GROUPS], how many GeeLark cloud phones do I have on each Android version and phone model? Put the results in a simple table.
3. Clean up and organize tags
As you use GeeLark, tags can pile up. Some may no longer be used, while others may have almost the same name. Ask AI to review your tags and the cloud phones they’re linked to, find unused tags, and suggest a cleaner naming system.
Prompt:
Review the tags for [all my cloud phones]. Find unused tags, similar names, and phones without tags. Suggest a cleaner naming system, but don’t make changes yet.
4. Install different apps across cloud phones
Say you’ve just created 50 cloud phones for different social media projects. One group needs TikTok and Instagram. Another needs TikTok, Facebook, and Reddit.
Tell the AI which apps each group needs. It can start the phones, install the apps, and shut the phones down afterward. You won’t have to switch between groups or set up every install yourself.
Prompt:
I’ve just created new cloud phones in [Group A] and [Group B], and they’re all turned off. Start the phones, install [TikTok and Instagram] on Group A and [TikTok, Facebook, and Reddit] on Group B. Once installation is complete, shut down all the phones and tell me the results.
5. Check and organize your proxies
If you’ve imported a lot of proxies into GeeLark, you may forget which ones still work, which ones are assigned to phones, and which ones aren’t being used.
Ask GeeLark MCP to check their connection status, see which phones use each proxy, and find unused proxies. The AI can also suggest how to group them by country or project.
Prompt:
Check all my proxies in GeeLark. Find any that fail connection checks or aren’t assigned to a cloud phone. Show me which phones are using each proxy and suggest how to organize them by [country or project]. Don’t delete or change anything yet.
6. Warm up TikTok accounts
Say you manage multiple TikTok accounts and want to use GeeLark’s TikTok AI account warmup template to schedule daily browsing and warmup tasks.
You can ask AI to plan the tasks and create them in GeeLark. Even with 100 or 200 accounts, you don’t have to configure each task by hand.
For example, AI can choose search topics, video watch time, whether to view comments, and the chances of liking or saving videos based on each account’s niche. It can also set different dates and times for different cloud phones.
Prompt:
I’ve got 10 TikTok accounts in the [TikTok demo] group. Set up three days of tasks for them using GeeLark’s TikTok AI account warmup template. Give each account its own niche, with relevant search topics and different schedules based on local time. Once the tasks are created, show me the schedule and settings for each phone.
But don’t expect everything to be perfect after one prompt.
Your AI may not understand every setting in the template or what you want for each account. You may need to clarify the account niches, adjust the settings, and correct mistakes over a few messages.
This takes some work at first. Once the AI understands your needs in the current chat, it can reuse the rules you’ve agreed on for similar tasks.
Here’s an example of TikTok warmup tasks I created with GeeLark MCP.
7. Schedule TikTok and Instagram Reels posts
If you post short videos across many TikTok and Instagram accounts, AI can also help you organize content and schedule posts.
Give AI your videos and some details about the content. It can draft captions, then use GeeLark MCP with the TikTok video posting and Post Reels video on Instagram templates to assign videos, fill in captions, set options, and schedule posts.
Prompt:
I’ve got videos ready in [your computer folder]. Schedule them for my TikTok accounts in [Group A] and Instagram accounts in [Group B] over the next [7 days]. Upload the videos as needed, write captions that fit each video and platform, and schedule posts at different times based on each account’s local time. Once everything is set up, show me the posting schedule and captions.
Tips for using GeeLark MCP
After spending some time with GeeLark MCP, I’ve picked up a few lessons that may help if you’re just getting started.
1. Let AI get to know your GeeLark workspace
When you first connect GeeLark MCP, start with a few simple questions.
Ask how many groups you have, which cloud phones are in each group, which tags and proxies they use, and whether any proxies are shared by several phones.
This gives AI a basic picture of your workspace. Later, when you ask it to organize devices or run bulk tasks, you won’t have to explain everything from scratch.
2. When AI gets confused, send it back to the docs
AI won’t always understand GeeLark MCP correctly. It might choose the wrong tool, use the wrong settings, or say something can’t be done when it can.
When that happens, I ask it to review the official GeeLark CLI documentation, check the available tools and settings, and try again.
Some tasks also take more than one prompt. You may need to explain what you want, point out mistakes, and go back and forth until AI gets it right.
3. Different AI agents may behave differently
In my tests, two AI tools sometimes responded very differently to the same task.
For example, Codex sometimes refused to schedule certain social media automation tasks, even when I only asked it to set up an existing GeeLark template.
If this happens, try being more specific. Name the template, explain which settings you need, and say whether you want the task created. Still, changing the prompt won’t always solve an unsupported action or an AI tool’s limits.
With OpenCode and DeepSeek V4, I haven’t run into the same issue so far. That’s only my experience, though. You can try different AI agents to see which fits your workflow best.
4. Explore other MCP tools for Android automation
GeeLark MCP mainly gives AI access to GeeLark’s built-in features. If you want AI to interact with Android screens, such as tapping buttons, typing text, swiping, or testing apps, take a look at this open-source project:
It uses ADB and uiautomator2 to let AI interact with Android devices.
Keep in mind that this is a third-party project, not part of GeeLark’s official MCP server. Check the ADB connection requirements and test whether it works with your GeeLark cloud phones.
5. Make sure cloud phones are turned off when a task ends
Starting a cloud phone through GeeLark MCP is different from opening one by hand. AI can start and control the phone in the background, so you may not see a phone window appear on your computer.
If your instructions aren’t clear, or AI misunderstands them, it might leave phones running after the task ends. If you don’t notice, those phones can keep using paid minutes.
For tasks that don’t need phones to stay on, such as installing apps or changing device settings, add one clear instruction: turn off every phone you started when the task is done, and confirm that it’s off.
That can help you avoid paying for time you didn’t mean to use.
Start using GeeLark MCP
GeeLark MCP’s biggest benefit is that AI can do more than tell you how to use GeeLark. It can carry out tasks for you.
From creating cloud phones and organizing proxies to installing apps in bulk and scheduling automation, you can describe what you want and let AI call the right GeeLark tools.
If you’re new to MCP, start small. Ask AI to count your cloud phones or check your proxies. Once you’re comfortable, move on to bulk actions and automation.
AI won’t always get it right on the first try. When you change lots of devices or create tasks, check the results and keep refining how you work with it.