MCP app for Android, drive apps via AI (no root, PII redacted locally)
Hi HN, author here. Android Remote Control MCP is an MCP server that runs directly on your Android phone (no root, no ADB, no computer in the middle) and lets an AI agent drive real apps the way a human would: it reads the screen through the accessibility tree, taps, types, scrolls, optionally also screenshots. I spent a lot of time optimizing tool usage and token consumption, and making it work not only via local harnesses but also via Claude.ai / Claude Desktop and chatgpt.com (if you have the proper account, the app acts as its own OAuth server, you approve connections with a code on…
What it does
An MCP Server for Android running on the phone, optmized for token usage, supports also files downloads and cloudflare and ngrok automated tunnelling. - danielealbano/android-remote-control-mcp
In the maker’s words, at launch
Hi HN, author here. Android Remote Control MCP is an MCP server that runs directly on your Android phone (no root, no ADB, no computer in the middle) and lets an AI agent drive real apps the way a human would: it reads the screen through the accessibility tree, taps, types, scrolls, optionally also screenshots. I spent a lot of time optimizing tool usage and token consumption, and making it work not only via local harnesses but also via Claude.ai / Claude Desktop and chatgpt.com (if you have the proper account, the app acts as its own OAuth server, you approve connections with a code on the phone). The newest part is Privacy Mode, and it exists because after an earlier release someone told me in plain terms they'd never use it because, rightfully, they didn't want the LLM provider to see everything on their screen! So now a combination of a small local model plus deterministic detectors identify personal information (emails, phone numbers, credit cards, IBANs, national IDs, English names, etc.) and redact it on-device before anything leaves the phone, with a benchmarked detection rate of about 87% (the benchmark is in the repo; non-English names, are the current weak spot but I am working on it). Why did I build it? I want my agents to be able to use my phone to do searches, book things for me and do sometimes boring stuff ... without sharing all my data with service providers and without having to run a local LLM! Tradeoffs: the service declares itself as an accessibility app so it can read apps which also makes it impossible to distribute this app on the Google Play store, hence it's distributed via GitHub at the moment (there's a standard build and a FOSS build without Google Play Services which soon will be published on F-Droid). Also, because I care about prompt injection as much as privacy, the data returned to the agent is prefixed with a message, which uses quite some strong wording, to make it clear that the content is supplied by third party sources and is an untrusted input! Mitigates the potential attacks a lot. Is it perfect? Nope, there's plenty to improve, especially around apps coverage but even small models (like Haiku) can drive the phone and handle unseen situations if you give them enough detail! What I am working on next? Three major things: - a free reverse tunnel, encrypted end to end, with Let's Encrypt certificates to be able to use the app with Claude.ai / Claude Desktop / Chatgpt.com and such for free and with much more privacy than using ngrok or cloudflare (which don't provide e2e encryption but only encryption to the edge, so they see everything) - a skills database for the apps, so your agent can work out what to do much faster - a custom model to improve the privacy standing, especially for non-English names And, additionally, a new guided setup and a revised UI to make it easier to start to use the app!
Does the same job
all alternatives →
- WMWhatsApp MCP Server2025 · github.com · ▲229
Hi HN – I built an open-source, self-hosted Model Context Protocol (MCP) server for WhatsApp: https://github.com/lharries/whatsapp-mcp It connects to your personal WhatsApp account via the WhatsApp Web multi-device API (using whatsmeow from the Beeper team), and doesn't rely on third-party APIs. All messages are stored locally in SQLite. Nothing is sent to the cloud unless you explicitly allow your LLM to access the data via tools – so you maintain full control and privacy. The MCP server can: - Search your messages, contacts, and groups - Send WhatsApp messages to…
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 27d ago · cactuscompute.com
Launched alongside, August 2026
the whole month →- TL
Life & fun · 10d ago · louisabraham.github.io
- SA
Hello HN! I found that picking out plausible but diverse skin tones for my digital art and game development projects was kind of difficult, and I got curious about if there was a way to define a color space that made it easy. I've built a color picker and procedural generation algorithm based on the space as well as a bunch of other fun js features and demos throughout the page that use the equations. If you find it interesting, I have lots of explanations of how I built it and what properties the space has. The methodology might be a bit shaky, but hopefully the result is as helpful for…
Life & fun · Aug 2026 · toneyalexander.github.io
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com