Tech & AI Writer
Alex Rivera
Alex covers the intersection of AI tooling, developer ergonomics, and consumer software. Particularly interested in how the Model Context Protocol is reshaping what apps can be in an AI-native era.
Posts by Alex (13)
I asked Claude what to order, then told it to log it
Macroline runs as an MCP server, so an AI assistant can read your diary and write to it. Here's a real workflow: ask Claude for the best thing to order, then have it log the meal, without the AI ever inventing a number.
What we built — Macroline 1.0 ships today
Today Macroline 1.0 lands on the App Store. Here's what's in the box, what we deliberately left out, who 1.0 is for, and the design principles behind the cuts. Substance, no hype — that's the brand.
No tracking, no theater — the macroline.app stack, in full
Most consent-banner sites set thirty or more third-party cookies on first load. macroline.app sets zero non-strictly-necessary cookies. Here's what we actually use, what we deliberately don't, and why the popular "Accept all / Reject" pattern is theater more often than it's a defensible privacy practice.
Why we built an MCP server for nutrition data
Macroline runs as a Model Context Protocol server. Here's what that means, why it matters, and how AI-native tools change what a tracker can be.
AI-native vs. AI-bolted-on — a developer's take
The difference isn't a chatbot button. It's whether the product was designed assuming an AI agent would be a primary user. Most "AI features" still aren't. Here's how to tell, and why it matters.
Building an MCP server — lessons from the trenches
Notes from implementing a production-ready MCP server with OAuth 2.1, dynamic client registration, and a real consumer app on the other end. The places we got it wrong, and how the spec helps and hurts.
Designing for AI clients vs. human clients — same data, different shape
A REST API for humans and an MCP server for AI agents look superficially similar. They aren't. The contracts, error semantics, and data shapes that work for one are quietly bad for the other. Here's what the second-class consumer of your API actually needs.
How MCP changes the API economy
The Model Context Protocol isn't just another integration spec — it's a shift in who the primary user of an API actually is. Here's what that means for product builders.
When AI is wrong about food (and how to spot it)
AI is good enough at parsing meals into macros that the failure modes are no longer obvious. The wrong number doesn't look wrong. Here's a taxonomy of the specific ways AI gets food data wrong, and what to look for in your own logs.
Why I don't connect my food log to my AI assistant
I work on AI tooling. I build MCP servers. And I deliberately do not connect my personal nutrition data to a cloud AI assistant. Here's the threat model, what 'connected' actually exposes, and the version of this I do use.
MCP turned one — what shipped, what didn't
Anthropic announced the Model Context Protocol in late 2024. A year and a quarter in, the protocol has stabilized in surprising ways and stalled in others. Here's what's actually in production, what got built and abandoned, and where I think the next year goes.
Tools we wish existed in the macro-tracking category
An honest builder's wishlist for what's still missing in nutrition tracking, from substitution preview to cross-tracker import. Open product brief, not a roadmap.
Why we wrote our own nutrition parser instead of stuffing GPT into it
A vanilla LLM call looks like a free meal parser until you ship it. Here's what breaks (units, portions, brand canonicalization) and why a structured parser with confidence scoring is the right tool.