I Still Prefer MCP over Skills

Comment section explodes: MCP loyalists vs “Skills” truthers

TLDR: Author says Model Context Protocol—a connector for AI apps—beats “Skills” for real integrations. Comments erupt: critics say skills can call APIs too, pragmatists split the difference, and jokers frame it as hammer vs screwdriver, making this a use‑the‑right‑tool moment that matters for future AI workflows

MCP vs Skills just turned into the internet’s latest tools cage match. The author of the post declares love for MCP—think a plug-in gateway that lets AI talk to your apps—saying “Skills” (how-to bundles) become a mess when they need command-line tools. He touts MCP’s no-install, auto-updates, easier sign-in, and works-anywhere vibe. Then the comments hit turbo.

The loudest clapback? “Skills can call APIs too.” User charcircuit blasts the take as a strawman, saying the CLI complaint “invalidates the entire article.” On the other side, robotobos drops nuance: keep Skills for teaching style and one-off know‑how, but use MCP for repeatable chores; he even hard-coded a script into a tool to make runs faster and consistent. Meanwhile, ghm2199 flexes a DIY win: a home-server MCP so his AI can reach files from any device—“works on my laptop, phone, and web,” hero moment unlocked.

Then the peacemakers arrive. Alierfan argues it’s not a zero-sum game: MCP is the infrastructure, Skills are the playbook. And meme-lord leonidasv nails the mood with “hammer vs screwdriver.” Translation for the non-nerds: this is less a deathmatch and more a use-the-right-tool debate—with plenty of spice, a few bruised egos, and a whole lot of “skill issue” jokes.

Key Points

  • The article argues MCP is a superior architecture to Skills for granting LLMs access to services due to its API abstraction model.
  • MCP’s cited advantages include zero-install remote usage, seamless updates, OAuth-based authentication, portability across clients/devices, sandboxing, tool discovery, and auto-updates.
  • Skills are described as effective for encoding knowledge or conventions but problematic when they depend on local CLIs for execution.
  • Many common LLM clients (e.g., standard web interfaces) cannot run CLIs, limiting the practicality of CLI-dependent Skills to compute-enabled environments.
  • CLI-based Skills introduce deployment, distribution, and secret-management challenges, particularly in ephemeral environments.

Hottest takes

"MCP’s are great for custom, repeatable tasks." — robotobos
"invalidates the entire article." — charcircuit
"This is the same as saying \"I still prefer hammer over screwdriver\"." — leonidasv
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.