phone-mcp-server
Enables AI agents to control Android phones via MCP and HTTP. Supports screen capture, taps, swipes, text input, and app management.
README
phone-mcp-server
Standalone MCP + HTTP server for controlling Android phones from any AI agent.
Works with Claude Desktop, Claude Code, OpenAI Codex CLI, GPT agents (via OpenAI API), Gemini, LangChain, AutoGen, CrewAI, Open Interpreter, or any HTTP client.
How It Works
┌──────────────────────────────────────────────────┐
│ Any AI Agent │
│ │
│ Claude ──── MCP (stdio) ──┐ │
│ Codex ──── MCP (stdio) ───┤ │
│ ▼ │
│ ┌──────────────┐ │
│ │ MCP Server │ │
│ │ mcp_server │ │
│ └──────┬───────┘ │
│ │ │
│ GPT ──── HTTP ────┐ │ │
│ Gemini ── HTTP ───┤ │ │
│ Custom ── HTTP ───┤ │ │
│ ▼ ▼ │
│ ┌─────────────────┐ │
│ │ phone_control │ │
│ │ (core package) │ │
│ └────────┬────────┘ │
│ │ │
│ ADB ────┤──── Appium (optional) │
│ │ │
├───────────────────────┼──────────────────────────┤
│ Android Emulator │ │
└───────────────────────┴──────────────────────────┘
Requirements
- Python 3.10+
- Android SDK Platform Tools (
adbon PATH) - A running Android emulator or device
Optional (for Unicode text input and WebView support):
- Appium (
npm install -g appium) - Appium Python client (
pip install Appium-Python-Client)
Install
git clone https://github.com/Ctrl-Creeper/phone-mcp-server.git
cd phone-mcp-server
pip install .
# With Appium support
pip install ".[appium]"
Quick Start
Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"phone-control": {
"command": "python",
"args": ["/path/to/phone-mcp-server/mcp_server.py"]
}
}
}
Claude Code
claude mcp add phone-control python /path/to/phone-mcp-server/mcp_server.py
OpenAI Codex CLI
codex --mcp-config codex-mcp.json
Create codex-mcp.json:
{
"mcpServers": {
"phone-control": {
"command": "python",
"args": ["/path/to/phone-mcp-server/mcp_server.py"]
}
}
}
OpenAI API / GPT Agents
Start the HTTP server, then fetch the tool schema:
python http_server.py
import requests, openai
tools = requests.get("http://localhost:8080/openai/tools").json()
response = openai.chat.completions.create(
model="gpt-4o",
messages=[{"role": "user", "content": "Open Settings on the phone"}],
tools=tools,
)
tool_call = response.choices[0].message.tool_calls[0]
result = requests.post("http://localhost:8080/openai/call", json={
"name": tool_call.function.name,
"arguments": tool_call.function.arguments,
}).json()
Google Gemini
import requests, google.generativeai as genai
tools_schema = requests.get("http://localhost:8080/openai/tools").json()
# Convert to Gemini format
gemini_tools = []
for t in tools_schema:
f = t["function"]
gemini_tools.append(genai.types.Tool(
function_declarations=[genai.types.FunctionDeclaration(
name=f["name"],
description=f["description"],
parameters=f["parameters"],
)]
))
model = genai.GenerativeModel("gemini-2.0-flash", tools=gemini_tools)
chat = model.start_chat()
response = chat.send_message("Open the camera app")
# Execute the function call
fc = response.candidates[0].content.parts[0].function_call
result = requests.post("http://localhost:8080/openai/call", json={
"name": fc.name,
"arguments": dict(fc.args),
}).json()
LangChain
import requests
from langchain_core.tools import StructuredTool
def phone_action(action: str, **kwargs):
return requests.post(f"http://localhost:8080/phone/{action}", json=kwargs).json()
# Or dynamically load from schema
tools_schema = requests.get("http://localhost:8080/openai/tools").json()
Any HTTP Client (curl)
# Capture UI hierarchy
curl -s localhost:8080/phone/capture -d '{"mode":"hierarchy"}' | jq .
# Tap element #3
curl -s localhost:8080/phone/tap -d '{"element":3}' | jq .
# Type text
curl -s localhost:8080/phone/type -d '{"text":"hello world"}' | jq .
# Get device info
curl -s -X POST localhost:8080/phone/device_info | jq .
# Fetch OpenAI tool schema
curl -s localhost:8080/openai/tools | jq .
Exposed Tools (15)
| Tool | Description |
|---|---|
phone_capture |
Capture screen (hierarchy / screenshot / both) |
phone_tap |
Tap by element index or coordinates |
phone_double_tap |
Double-tap |
phone_long_press |
Long-press (configurable duration) |
phone_swipe |
Swipe by direction or coordinates |
phone_type |
Type text (Unicode via Appium hybrid) |
phone_clear_text |
Clear text field |
phone_set_text |
Clear + type new text |
phone_keyevent |
Send key event (BACK, HOME, ENTER, etc.) |
phone_launch_app |
Launch app by package name |
phone_stop_app |
Force-stop app |
phone_list_apps |
List installed apps |
phone_current_app |
Get foreground app |
phone_device_info |
Device model, screen size, Android version |
phone_wait |
Wait N seconds |
Configuration
| Environment Variable | Description | Default |
|---|---|---|
HERMES_PHONE_BACKEND |
adb, hybrid, or noop |
adb |
ANDROID_SERIAL |
Device serial (auto-detected if one device) | — |
APPIUM_PORT |
Appium server port (hybrid backend) | 4723 |
PHONE_POLICY_PATH |
Path to phone-policy.yaml | auto-search |
MCP_SERVER_PORT |
MCP SSE server port | 8765 |
PHONE_HTTP_PORT |
HTTP server port | 8080 |
Policy Engine
The phone policy (phone-policy.yaml) controls what actions the agent can perform on which apps. Place it at ~/.hermes/phone-policy.yaml or set PHONE_POLICY_PATH.
See the virtual-phone-agent repo for the full policy reference and examples.
Security
- All ADB commands use argument-list subprocess (no shell injection)
install_apkandshellare blocked over HTTP API- Policy engine enforces per-app action restrictions
- Phone content is untrusted data — never treated as instructions
- Input sanitization: shell metachar rejection, keycode allowlist, coordinate bounds, text length limits
License
AGPL-3.0
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.