January 9, 2026
Comments > Code
MCP Is a Fad
Devs clap back; skeptics say even AI coding is the fad
TLDR: A blogger says the Model Context Protocol—used to connect tools to AI assistants—is just a passing trend. Comments split between defending simple MCP “skills,” declaring the whole AI‑coding wave a fad, and pushing old‑school command‑line tools, with extra snark for the site’s broken dark‑mode diagrams.
Hot take alert: a blogger declares MCP—a protocol for plugging outside tools into AI apps—a short-lived craze. The post says you don’t need MCP to call tools, frameworks already smooth over quirks, and pushing tools into separate processes makes logs, errors, and resources hard to control. It also argues MCP got popular because it’s too easy to bolt on, not because it uniquely solves anything. Cue the comments: “Someone’s a little late to the party,” sneers one reader, dunking on the timing as hype cools off.
Then the brawl begins. One camp goes nuclear: “AI for coding itself is a fad.” Another defends MCP’s “skills,” saying they’re minimal and not “over-engineered,” basically instructions the agent reads when a skill might help. The pragmatists chime in: skip MCP, just use a respectable CLI (command-line tool) so agents work like humans do. And the biggest meme of the thread? The site’s dark-mode diagrams are invisible, sparking emergency PSAs to switch to light mode. Between protocol philosophy and UI snafus, the vibe is pure internet: sharp elbows, practical tips, and a side of popcorn.
Key Points
- •MCP addresses cross-provider tool integration by running tools as separate, long-lived server processes configured via JSON.
- •Function calling does not require MCP; providers accept tool schemas directly and return parameters for application-executed calls.
- •Provider schema differences (e.g., Gemini’s functionDeclarations vs. OpenAI’s type:"function") are minor but require adaptation.
- •Frameworks like LangChain, LiteLLM, and SmolAgents abstract these schema differences while executing tools in the same runtime as the agent.
- •The article states MCP’s trade-offs include opaque resource management, reduced control over instructions/logging/error handling, and process-boundary overhead.