Nourishing technology: why I built an MCP server to connect AI with my running data
We shape our environments, and thereafter they shape us. Great technology does more than solve problems. It weaves itself into the world we inhabit. At its best, it can expand our capacity, our connectedness, our sense of what’s possible. Technology can bring out the best in us.
It’s 5:30am and I’m walking home from the gym, legs a little jiggly. Watching the sky shift from magenta to orange to blue. I take a quick moment to track the workout and tag a few usual partners in crime, and move on feeling more ready for my day ahead.
This is what technology can feel like. Ambient. Warm. In service of something real.
But we all know that feeling isn’t guaranteed. I quit Instagram last year. And the tipping point was when I counted the number of ads compared to real content from people I cared about, it was over 50% ads. They were perfectly targeted to my taste on stuff I didn’t need but suddenly wanted. The loops that went on in my head were terrifyingly close to when the poor woman with brain implant started reciting ads involuntarily from Black Mirror. I was spending mental energy every day fighting an algorithm that knew me well enough to find the gaps in my defenses and exploit them.
The technology knew me. It just wasn’t on my side.
Research1 links fitness tracking to disordered eating, body anxiety, compulsive exercise, especially for those who are already vulnerable. The technology that’s supposed to help us get healthier often leaves us more stressed, more distracted, and further from our own realities.
Technology echoes whatever is living within us.
One mile at a time
I’d been at the gym five days a week for a few years. Then my friends talked me into signing up for a marathon.
I had no idea how to train for one. The running world had its own language: tempo runs, lactate threshold, negative splits. Training plans assumed I already knew what I was doing.
What got me started was my cross-training buddies suggesting we run a mile together after class. Just one mile. Then one became two. Two became three. The mileage built because the people made it feel natural.
There’s a guy at my gym, Carlo (not his real name), who shows up every single day. He is often the class clown, always organizing something, and the social glue. If I miss class for a few days, he’ll text to check if I’m ok and makes it a big deal in a hilarious way. In a world where we’re becoming more isolated and self-contained, someone caring enough to notice feels precious.
Strava became where that encouragement lived between workouts. A kudos from a gym friend is the same high five they’d give me in person, and posting my run might be the thing that gets them out the door tomorrow. Watching friends log their training week after week made consistency feel contagious. The platform held space for something that was already real.
The first time I ran ten miles, I didn’t feel like I’d broken through anything. I felt like I’d been carried there — by the people who showed up, by the miles that came before, by the accumulation of many early mornings.
I finished that marathon. Now I run one every year.
We can often tell contrasts in technology that expands our capacity with platforms that drain “the depth and warmth from everything they touch.”
The difference comes down to grounding with reality vs. disconnect. The first allows us to connect with our surroundings and feel like we are part of something greater, while the latter results in fear and anxiety because there is no coming on top of illusions of perfection. Scrolling strangers’ highlight reels triggers inadequacy because we measure ourselves against curated performance and come up short. But with seeing a friend’s workout and leaving a comment in admiration, we’re participating in each other’s lives. The same underlying feature of showing what others are doing produces opposite psychological effects.
What would personal AI look like if they followed this principle? Can we design intelligence that promotes our growth and long-term wellbeing and break out of seconds-long dopamine feedback loops holding our attention hostage?
I started building a Model Context Protocol (MCP) server connecting Claude to my Strava data. MCP is an open standard that lets AI agents interface with external tools and data sources. Instead of building custom integrations for every combination of AI and data source, you build once and connect anywhere.
When I asked “Based on my Strava stats, can you suggest a plan to make me run faster? I have time to run about 20 miles a week.”, Claude asked follow-up questions because it knew I always want to clarify assumptions. What’s your current pace? Do you have any injuries? What days work best? We went back and forth, adjusting and refining until we landed on something that actually fit my life.
It felt like working with a coach: one who patiently met me where I was at every decision point. I didn’t need to already know what I was doing. The understanding built through conversation.
Here are a couple of other questions I’ve asked:
“I’m traveling to San Diego and staying at Mission Beach—find me some good running routes on Strava.”
“Make me a training plan for this week based on the weather in Seattle. I don’t want to run in the rain or when it’s too cold.”
The magic of natural language interactions is that we unlock functionality and flexibility that static buttons and dashboards can’t anticipate. Software starts to become more liquid.2 With that fluidity, data crosses the boundaries of individual apps - health tracking, meeting schedules, journal notes, etc. can now come together from previous silos to form a more complete personal context than ever before.3
This evolution towards holistic context and infrastructure for personal AI ecosystems give rise to assistants that are genuinely ours. They can inspire us to do more, and make the intimidating feel possible.
Technology that is for us
We’ve stumbled into something incredibly powerful with LLMs. The iteration loop between idea and working software has collapsed. You can describe what you want and have it exist. Software becomes fluid. Personal.
The strava-mcp server connects my data to an assistant that actually belongs to me, that I can shape and change as I learn what I need. I don’t have to wait for a product team to prioritize my feature request. I don’t have to hope the metrics align with what would actually help me. I can just try something, see if it works, and adjust.
The question now becomes whether we will continue to build more of what captures attention, or technology that actually cares for us. Technology that is warm like a trusted friend. That knows you. That keeps your interest at heart.
The strava-mcp server is open source and available on GitHub. If you’re interested in connecting Claude to your own data sources, the MCP documentation is a good starting point.
Chen, Y., et al. (2024). “Exercise or lie down? The impact of fitness app use on users’ wellbeing.” Frontiers in Public Health.
Tzemou, E., et al. (2024). “The ‘Dark Side’ of General Health and Fitness-Related Self-Service Technologies: A Systematic Review.” Journal of Public Policy & Marketing.
Anderberg, I., et al. (2025). “The link between the use of diet and fitness monitoring apps, body image and disordered eating symptomology: A systematic review.” Body Image.
“Associations Between the Use of Fitness and Diet Tracking Technology and Disordered Eating Behaviour: A Systematic Review.” (2025). European Eating Disorders Review.
Alex Komoroske talks a lot about infinite software in his shared thoughts Bits and Bobs.
Daniel Miessler does a great breakdown of how he is thinking about and building personal AI infrastructure.