droid-mcp
Let an AI use your Android phone: an app and an SDK with 160+ MCP tools (calendar, SMS, files, camera, notifications, screen control) for Claude Code, Codex, Cursor or on-device models, with approvals on the phone.
Install / Use
claude mcp add stixez -- npx -y github:stixez/droid-mcpIf the server publishes to npm under a different name, use that package instead — check the repo README.
MCP Server
Model Context Protocol server
Quality Score
Category
AutomationSupported Platforms
Tags
Our assessment of droid-mcp
droid-mcp scores 84/100 on our quality scale, 2022nd of 2,790 Automation skills we index.
Its MCP Server is 7.4 KB long, split into 6 sections with 4 code examples: a thorough specification that gives an agent plenty to work with.
It has 45 GitHub stars, so there is little community track record yet; judge it on its content.
Maintenance, license and trust
- The repository was last updated today, so droid-mcp is actively maintained.
- It is released under the Apache-2.0 license, a permissive license that allows use, modification and commercial use with attribution.
- Its trust signals score 97/100, with no cautions. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.
Safety scan
No issues foundOur scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful.
AI review by kimi-k2.7-code on 2026-11-14. Automated pattern scan on 2026-11-14. It catches known dangerous patterns, not every risk — read a skill before letting an agent act on it.
droid-mcp compared with similar skills
All 4 of these similar skills score higher than droid-mcp; compare them before choosing.
| Skill | Score | Stars | Updated | Format |
|---|---|---|---|---|
| droid-mcp (this skill)by stixez | 84 | 45 | today | MCP Server |
| Agent-Reachby Panniantong | 100 | 112.3k | 8d ago | CLAUDE.md |
| rufloby ruvnet | 100 | 81.9k | today | CLAUDE.md |
| headroomby headroomlabs-ai | 100 | 79.8k | today | CLAUDE.md |
| CowAgentby zhayujie | 100 | 48.0k | today | CLAUDE.md |
Frequently asked questions
- How do I install droid-mcp?
- Run
claude mcp add stixez -- npx -y github:stixez/droid-mcp. The install tabs above show the steps for each supported agent. - Which AI agents does droid-mcp work with?
- It is written for Claude Code, Claude Desktop, Cursor and OpenAI Codex, as a MCP Server file. Other agents that read the same format can often use it too.
- Is droid-mcp safe to use?
- Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful. It is Apache-2.0-licensed and scores 97/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
- Is droid-mcp still maintained?
- The repository was last updated today, so droid-mcp is actively maintained.
Skill content
View source on GitHubThree ways to use it:
- The app: install it, allow what you're comfortable with, and connect Claude Code, Codex or Cursor. No code needed.
- SDK, on-device: your app runs a local model that calls the tools directly.
- SDK, MCP server: your app runs an HTTP server on the phone that any MCP client can reach over Wi-Fi.
There are 160+ tools in 50+ modules, each with only its own permissions. See all tools.
Use your phone from Claude Code
-
Install the app. Download
droid-mcp-<version>-arm64-v8a.apkfrom the latest release (older 32-bit phones:armeabi-v7a). Obtainium can keep it updated. -
Allow what clients may use. The setup screen explains each permission and how many tools it unlocks. Only notifications are needed.
-
Start the server and copy the command into a terminal on your computer:
claude mcp add --transport http phone http://192.168.1.20:8080/mcp --header "Authorization: Bearer <token>"The same card has config for Codex and JSON for other clients. Phone and computer must share a Wi-Fi, or use USB with
adb forward tcp:8080 tcp:8080.
Then ask "What's on my calendar today?"
<p align="center"> <img src="docs/images/ask-first.gif" width="820" alt="Claude Code is asked to text Mom; the phone shows an approval notification, the user taps Allow, the text appears in Messages, and Claude confirms it was sent" /> </p>The app asks you before a client sends, deletes or changes anything (once, for an hour, or always), logs every call, can give each computer its own token, and can pause itself off your home Wi-Fi. It isn't on Google Play, whose policies don't allow AI control through accessibility or SMS access.
<p align="center"> <img src="docs/images/home.png" width="260" alt="Home screen: server on, a client connected, and the command to copy" /> <img src="docs/images/point-and-wait.gif" width="260" alt="An AI agent points at the add-alarm button in the Clock app, the user taps it through the overlay, then taps Done" /> </p>For steps only you should do, the agent can point at the right spot on your screen and wait until you've done it (guide_user, the clip on the right). Your taps go through to the app underneath.
Use the SDK
// settings.gradle.kts
dependencyResolutionManagement {
repositories { maven { url = uri("https://jitpack.io") } }
}
// build.gradle.kts: pick modules, or droid-mcp-all for everything except the opt-in ones
dependencies {
implementation("com.github.stixez.droid-mcp:droid-mcp-core:0.14.0")
implementation("com.github.stixez.droid-mcp:droid-mcp-calendar:0.14.0")
}
Call tools from an on-device model:
val mcp = DroidMcp.builder()
.addTools(CalendarTools.all(context))
.build()
val toolsJson = mcp.listToolsJson() // hand this to the model
val result = mcp.callTool("read_calendar", mapOf("start_date" to "2026-10-01"))
Or serve them to desktop clients:
val mcp = DroidMcp.builder()
.addTools(CalendarTools.all(context))
.enableHttpServer(context = context) // port 8080, token generated
.build()
mcp.startServer()
val token = mcp.serverToken
Calls run off the main thread with a timeout. Beyond that, the builder can:
- ask the user before risky calls (
confirmToolCalls) - serve resources and prompts
- enable HTTPS (
enableTls) - go read-only (
readOnly = true)
Connecting clients covers tokens, sessions and TLS.
<details> <summary>Packaging error about duplicate <code>META-INF</code> files?</summary>android {
packaging {
resources {
excludes += setOf(
"META-INF/INDEX.LIST",
"META-INF/io.netty.versions.properties",
"META-INF/LICENSE.md",
"META-INF/NOTICE.md",
"META-INF/license/**",
"META-INF/versions/9/OSGI-INF/MANIFEST.MF", // droid-mcp-tls only
)
}
}
}
</details>
Modules
| Area | Modules |
|------|---------|
| Personal data | calendar contacts sms calllog media files downloads clipboard |
| Device | device screen settings sensors health biometric keyguard flashlight vibration audio usb wallpaper ringtone dnd |
| Connectivity | network wifi bluetooth telephony nfc location web |
| Camera & vision | camera screenshot qr mlkit |
| Apps & intents | apps alarms intent print tts playback |
| Notifications | notifications notifications-reply notification-watch |
| UI automation | accessibility ime overlay |
| Opt-in | shizuku root (shell tools), audit, tls, server-service |
Your app requests permissions; the library never does. Each module has requiredPermissions() and hasPermissions(context), and some need special access granted in Settings (which ones).
Security
The server needs a bearer token and checks Origin, sessions and body size. File tools stay in shared storage, and web tools refuse private addresses. It listens on every network the phone joins, so use trusted networks or TLS. See the security model. droid-mcp sends no analytics; ML Kit (mlkit, qr) reports usage to Google.
Requirements
Android 9 (API 28)+ at runtime; compileSdk 36+, AGP 8.9.1+ and Kotlin 2.3+ to build.
Documentation
| | | |---|---| | Tools | Every tool, its parameters and permissions | | Connecting clients | Tokens, sessions, mDNS, TLS | | Security | Threat model and what the host app owns | | Shell tools | Shizuku and root | | Versioning | How releases change things | | API reference | KDoc |
sample-app is a one-screen example: calling tools in-process, asking for a permission and serving over HTTP. app is the full host app, with a foreground service, approvals and every kind of special access. Contributions are welcome; for anything non-trivial, open an issue first. Licensed under Apache 2.0.
Related Skills
Agent-Reach
112.3kGive your AI agent eyes to see the entire internet. Read & search Twitter, Reddit, YouTube, GitHub, Bilibili, XiaoHongShu — one CLI, zero API fees.
ruflo
81.9k🌊 The original agent harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. Features adaptive memory, self-learning intelligence, federation, vector RAG integration, and native Claude Code / Codex / Hermes and many more Integrated
headroom
79.8kCompress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.
CowAgent
48.0kOpen-source personal AI assistant & Agent Harness. Plans tasks, runs tools and skills, self-evolves with memory and knowledge. Multi-agent, multi-model, multi-channel. Lightweight, extensible, one-line install.
Languages
Trust signals
From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.
