You’ve probably asked ChatGPT or Claude what time it is. And you’ve probably gotten a confidently wrong answer.
Here’s the dirty secret of the AI industry: 2.5 years after ChatGPT launched, the most advanced reasoning systems on Earth — models that can write code, pass the bar exam, and debate philosophy — still cannot tell you what day it is. Not won’t. Can’t.
The smartest tools humanity has ever built don’t know what time it is. And nobody at the labs seems to care.
Think about that for a second. You can ask Claude to analyze a legal contract, and it’ll nail it. Ask it whether a news article from yesterday is still relevant? It’s flying blind. Ask it to schedule something next Thursday? It’s guessing. The model has no temporal context — no sense of “now.” It’s frozen in its training data, permanently stuck in the past, like a brilliant historian who doesn’t know what year they’re living in.
Now, the fix is absurdly simple. One developer, Firas D., built a hosted endpoint called MCP Clock. It’s a clock. That’s it. It tells AI models what time it is. You plug it into Claude, Codex, or any MCP-enabled client, and suddenly your AI knows what day it is. Scheduling works. News queries work. Date-dependent reasoning works.
While developers are busy building elaborate agent frameworks and debating AGI timelines, the actual bottleneck to AI usefulness is a clock. A literal clock.
This is the pattern nobody talks about. The AI industry is obsessed with pushing the frontier — bigger models, longer contexts, more parameters. But the things actually holding back AI in production aren’t intelligence problems. They’re infrastructure problems. Dull, unsexy, obvious-in-retrospect infrastructure problems.
A model that can reason about quantum physics but can’t tell you if yesterday was Tuesday isn’t a model with a minor gap. It’s a model with a foundational blindness. And the labs shipped it anyway, because “it can tell time” doesn’t make for a good demo at a developer conference.
Here’s what should unsettle you: if the AI industry missed something this basic — something a $5 Casio watch gets right — what else is missing? What other mundane, foundational primitives are absent from your AI stack while you’re off chasing the next big capability?
The gap between what AI can do and what AI is useful for isn’t measured in IQ points. It’s measured in the boring stuff nobody wants to build.
MCP Clock is a tiny tool solving a tiny problem. But it’s a flashing signal that the AI infrastructure layer is full of holes. The models are ready. The plumbing isn’t. And until someone fixes the plumbing, your AI is a genius locked in a room with no windows.
So if you’re building with AI, stop asking “how smart is the model?” and start asking “what basic human context is it missing?” You might be surprised how often the answer is something a toddler handles without thinking.
The next breakthrough in AI won’t come from a bigger brain. It’ll come from someone finally giving the brain a clock.
FAQ
Q: Surely the labs know about this and have a reason for not including a clock?
A: They do know. The reason is that injecting real-time data into model context introduces non-determinism — the same prompt gives different answers depending on when you ask. Labs prioritize reproducible, benchmarkable behavior over practical usefulness. It's a research priority, not an engineering oversight. But for end users, it's still a glaring gap.
Q: What does MCP Clock actually do for me if I use Claude or Codex?
A: It gives your AI model temporal awareness. Any task that depends on knowing the current date — scheduling, news analysis, deadline calculations, age computations, expiry checks — becomes accurate instead of hallucinated. You plug the endpoint into any MCP-compatible client and your model stops guessing what day it is.
Q: Isn't this just a band-aid? Shouldn't the model itself handle temporal reasoning?
A: No — and this is the contrarian point. Models SHOULDN'T bake in the current time because it breaks reproducibility. The real failure isn't architectural; it's that the labs didn't ship a simple, official tool to bridge the gap. Leaving it to a random developer on GitHub for 2.5 years is the actual scandal here.