- Runs it
- No single steward
- Projects
- 20
- Official
- 0
- Stars checked
- 2026-11-08
Mobile AI Agents and Phone MCP Servers: 20 open-source projects compared
Let an AI agent tap, type and test on real Android and iOS devices. Compare 20 open-source mobile agents and mobile MCP servers by stars, licence and activity.
Compare Mobile Agents projects
These 20 open-source projects implement or support Mobile Agents. All of them are community projects. Sort by stars or recent activity, or filter by type. These projects are listed from public GitHub data and are not scanned by AgenticMart.
OpenClaw
Personal AI assistant that runs across operating systems and chat channels
NanoClaw
Container-based assistant that connects to WhatsApp, Telegram and Slack
Open-AutoGLM
Open phone agent model and framework from Z.ai
Mika
On-device Android agent that operates apps and the web
OmniBot
Android app that runs an AI agent directly on the phone
OpenGUI
Android GUI agent framework that sees and operates app screens
droidclaw
Agent that controls an Android phone from a plain-English goal
PhoneClaw
Local phone agent runtime with on-device models and mobile skills
OpenDroid
Autonomous self-planning AI agent for Android
PhoneAgent
Experimental agent that completes tasks across iPhone apps
MobileClaw
Android AI agent runtime for phone control and app automation
Zafiro
Bring-your-own-key Android agent with phone control and shell access
MobileUse
GUI agent that automates real Android devices
Stars and activity checked 2026-11-08. These projects are listed from public GitHub data and are not scanned by AgenticMart. Built one? List it free now, paid selling opens soon.
What is Mobile Agents?
Mobile agents are AI agents that run on a phone, control a phone, or act as personal assistants through messaging apps such as WhatsApp and Telegram. They typically read the screen and tap through apps, or connect to devices through tools such as ADB. This is a group of independent projects and not a single protocol.
Mobile agents vs MCP: how do they relate
MCP is a protocol for connecting agents to tools, and a mobile agent is a kind of product that may or may not use it. Mobile MCP and Pinky are examples that make a phone available to an outside agent through MCP. In that setup the phone control is a tool, and the reasoning happens in the agent. New to MCP? Read what an MCP server is.
Mobile Agents questions
What are mobile AI agents?
Mobile AI agents are AI systems that operate a phone or reach you through phone-based chat apps. Mobile agents are a category of independent projects, not a protocol or a standard. Some read the screen and tap through apps, and others run as assistants you message.
Mobile agents vs MCP: how do they relate?
MCP is a protocol for connecting agents to tools, and a mobile agent is a kind of product that may or may not use it. Mobile MCP and Pinky are examples that make a phone available to an outside agent through MCP. In that setup the phone control is a tool, and the reasoning happens in the agent.
How do I get started with an AI agent on my phone?
Choose a project, enable developer mode and USB debugging on an Android phone, and install ADB on your computer. You will also need access to a model, either a hosted API or your own deployment. MobileUse's README recommends using a test device and watching the phone while it runs.
Are mobile AI agents ready for everyday use?
Maturity varies by project, and several describe themselves as early or experimental as of November 2026. Pinky calls itself pre-alpha, and Open-AutoGLM says it is for research and learning only. Treat each project's README as the guide to its current state.
Is there a standard protocol for mobile agents?
No, mobile agents are a category rather than a protocol, and the projects listed below each choose their own approach. Some control Android through ADB, some read the accessibility tree and some expose a phone to other agents over MCP. MCP is a general tool connection here and not a mobile agent standard.
How do AI agents control a phone?
Many phone agents look at the screen, decide on the next action and then carry it out in a loop. Open-AutoGLM sends screenshots to a vision model and taps over ADB or HDC. Mobile MCP reads the accessibility tree first and falls back to screenshots and coordinates.
Can an AI agent control an iPhone?
Yes, some projects support iPhones, though setup and abilities differ from Android. Mobile MCP works with the iOS Simulator and with real devices connected over USB and trusted. Pinky needs iOS 17.4 or later, and on versions before iOS 27 it can only view, read text and open apps unless a helper is installed.