For two months, I haven't thought about toilet paper once. Not kidding. Every so often, a new entry appears in my calendar: "Knuspr delivery," complete with a time window and everything. I didn't create it. My AI did. A few days later, my groceries for the week are sitting at my front door. Milk, fruit, cleaning supplies, the cherry tomatoes I added to the list last week with a voice command — all there.
That sounds like science fiction. But it isn't. This has been my everyday life for about two months now, and I want to tell you how it works, what's gone wrong, and why I believe this is the future of shopping.
Why Knuspr — and why automate at all?
Quick context: I've been ordering my groceries from Knuspr for a while. Not because I'm too lazy to go to the supermarket (okay, maybe a little), but because I value efficiency, service, and simplicity. Knuspr delivers in time windows, the quality is solid, and I save myself the 45-minute loop around the store plus checkout queue.
But even online ordering started to annoy me. Every week: open the app, scroll through categories, click-click-click, build the cart, pick a delivery window, place the order. That's 15–20 minutes of repetitive work, every single week. For items I order 80% identically each time.
As someone who works in tech and advises on AI, my immediate reaction to this kind of thing is: this has to be automatable.
And then Knuspr did something recently that changed everything: they launched an official MCP server.
What is MCP — and why is it a game-changer?
MCP stands for Model Context Protocol. It sounds technical, but the core idea is simple: it's a standardized protocol that lets AI assistants connect to external services and take actions within them. Think of it as a universal language that allows your AI to talk to apps, shops, and services — not just generate text, but actually do things.
Here's what makes it significant: MCP isn't some niche experiment. Anthropic developed it and released it as an open standard. OpenAI adopted MCP in March 2025 — notably, a protocol created by a direct competitor. When OpenAI adopts a rival's protocol, that says everything about its relevance. The Verge put it well: "MCP has already taken the industry by storm."
So Knuspr supporting this standard isn't a gimmick. It's a direct integration with the exact infrastructure that ChatGPT, Claude, and every serious AI assistant is built on. That gives the whole thing staying power — it's not going to break with the next app update.
How automated grocery shopping works in practice
Now for the interesting part: what does this actually look like day to day?
I use OpenClaw as my AI agent. Through the MCP server, OpenClaw is directly connected to Knuspr and can browse products, manage the cart, and place orders. The agent knows my order history, which products I buy regularly, and how often.
Passive mode: OpenClaw independently tracks when I last ordered certain consumables. Toilet paper every three weeks? Dishwasher tablets every six weeks? Cleaning supplies monthly? The agent picks up on these patterns from my order history and adds items to the cart before I run out.
Active mode: When something comes to mind, I just say: "OpenClaw, add cherry tomatoes to my shopping list." Done. No switching apps, no scrolling, no searching.
Checkout: Once the cart is ready, the agent finds a suitable delivery window, places the order, and — this is the part that still gets me every time — creates a calendar entry with the delivery window. I don't have to think about anything. The groceries just appear.
In effect, I've turned my weekly grocery run from an active task into a passive event. Instead of "I need to go shopping," it's now "Oh right, delivery's coming Wednesday." That's a fundamentally different feeling.
The trust question — and why it's less scary than it sounds
I know what you're thinking: "You let an AI spend your money?!" That's the obvious reaction, and it's a fair one. Forrester identifies consumer trust as the central challenge for agentic commerce — AI-driven shopping. People are rightfully skeptical when a machine makes purchasing decisions autonomously.
But here's the key point most people miss: the AI has shopping access, not financial access.
My payment details live with Knuspr. The MCP server lets the agent navigate the shop like a regular user — search for products, add them to the cart, place an order. But the AI doesn't have my credit card details. It has no access to my bank account. It acts like a highly efficient personal assistant who handles the shopping for me, while payment runs through my existing Knuspr account.
That's an important architectural distinction. And honestly, it's exactly the right design for this stage of the technology.
The reality check — why 40% of AI projects will fail (and mine won't)
Now it gets interesting. Because I don't want to play the AI hype preacher who acts like agents will be handling everything by tomorrow.
Gartner predicts that over 40% of agentic AI projects will be cancelled by the end of 2027 — due to "escalating costs, unclear business value, or inadequate risk controls." Most of these projects are described as "early-stage experiments driven by hype and often misapplied."
That's a sobering number. And I think it's accurate — in an enterprise context. Large organizations launch AI agent projects with vague KPIs, endless alignment meetings, and scope that grows faster than the value delivered. Projects fail there because of complexity, not technology.
So why does my use case work?
Three reasons:
- The ROI is immediately tangible. No quarterly reports needed. Groceries show up at the door, I saved time. Done.
- The stakes are low. Worst case: the wrong yogurt gets ordered. No business damage, no reputational risk. Just a mildly annoying fridge situation.
- The feedback loop is fast. If something weird gets ordered, I fix it right away. Over two months, I've had to course-correct a few times when odd things ended up in the cart. But no financial damage, nothing egregious. The AI learns quickly.
This is the honest version of AI adoption: not perfect from day one, but immediately useful and getting better fast.
Where this is all heading
My Knuspr setup isn't some isolated tech enthusiast's playground. What's happening here is the beginning of a fundamental shift in commerce.
Google unveiled its Universal Commerce Protocol (UCP) in January 2026 — developed together with Target, Walmart, and Etsy. When the world's biggest retailers are working with Google on a protocol that lets AI agents shop, this is no longer an experiment. It's strategy.
Groceries are just the most obvious first use case, because shopping here is especially repetitive and predictable. My weekly order isn't really a creative decision — it's logistics. And AI agents are logistics machines.
But think further: office supplies that reorder themselves. Pet food that never runs out. Household essentials that just exist. The line between "shopping" and "managing your household inventory" starts to blur — and AI agents will be the ones erasing it.
My honest verdict after two months
Two months of automated grocery shopping. Here's the straightforward assessment:
- Time saved: Noticeably significant. Not just skipping the supermarket trip (Knuspr had already handled that), but also the 15–20 minutes of weekly cart-building.
- Error rate: Low. A few odd product choices at the start that needed correcting. Nothing dramatic.
- Financial damage: Zero. Genuinely zero.
- Convenience gained: Enormous. Not having to think about shopping anymore is the real luxury.
Is it perfect? No. Does it take a bit of time upfront to teach the agent your preferences? Yes. But it pays off — and quickly.
If you're already using Knuspr and want to try this yourself, check out Knuspr's MCP server. And if you want to understand how to put AI agents to practical use in your daily life or business — not as a hype project, but as a genuine way to reduce friction — get in touch with me. These kinds of pragmatic use cases are exactly what I advise individuals and companies on.
The future of shopping isn't better apps. It's not having to open an app at all.