Back when I was first introduced to computers, we had real, physical control over our data. Ejecting a disk instantly removed access, and the motion made the separation feel real and inherently trustworthy.
In today’s cloud and AI world, that feeling has disappeared. There’s no clean way to grant temporary access, which forces users into an uncomfortable all-or-nothing trust model.
The Hackathon Spark
In a recent hackathon, our team tackled this challenge head-on, creating a secure method to access data from your phone and expose it as an MCP tool for AI clients such as ChatGPT, Claude, and Gemini.
How It Worked
Using the smashDATA app in ChatGPT, and the smashDATA app running on our phones, users were given full control over which data sources to share and how long they were shared.
Users simply opened the iOS app and selected the data to share (in our case Apple HealthKit workout data). Then, using a laptop or desktop, the user would chat with ChatGPT or Claude and request access to workout data. To the AI, the workout data was treated like any other tool, so a simple request of “get my workouts” pulled back a historical dataset, fully configurable via the iOS app.
Animated Architecture Overview
Why the MCP Model Matters
The real breakthrough behind this experience is the Model Context Protocol (MCP). MCP shifts the traditional data-access paradigm by putting the user in total control. Instead of piping data to cloud services or permanently integrating accounts, MCP works like a permissioned toolbelt for AI models.
- The model never stores your data.
- The user decides exactly which sources are exposed.
- Access exists only while the local server is active.
- Revoking access is as simple as closing the app.
This mirrors the physicality we lost during the cloud transition. Just as pulling a disk from a drive severed access, stopping the MCP server instantly cuts the AI off from your private data.
Exploring the Data
With 100 of my recent bike rides as context, I explored my data using ChatGPT voice mode. I pulled in weather conditions, chatted through patterns, and visualized trends I had never seen before.
When I was finished, I closed the app and the data instantly disappeared from AI’s reach.
It felt like ejecting that old disk again… but in a very different era.
The Broader Implications
This single use case example acts as a proxy for something much larger.
Our most valuable personal datasets remain locked away on our devices, siloed behind vendor ecosystems. Health metrics, photos, locations, financial records, messages, sleep data, environmental sensors, and more are trapped inside our phones with no safe, granular way to temporarily lend them to AI systems.
This is a future where trust is earned through architecture, not terms of service.
Artifacts
Workout Dashboard: https://www.smashdata.io/workouts/
Architecture Overview: https://www.smashdata.io/phonemcp/
