A blog post titled "I still prefer MCP over skills" sparked significant debate in early 2026 when it hit Hacker News front page with 319 points and 269 comments on April 10, 2026. The discussion clarified that Claude's Skills feature and Model Context Protocol (MCP) serve complementary rather than competing roles in AI development.
Progressive Discovery Eliminated Key Efficiency Gap
Early 2026 brought a crucial technical shift when MCP adopted progressive discovery in January 2026. This change mirrored the efficiency technique that made Skills context-efficient, reducing MCP tool descriptions from potentially 50,000+ tokens to compact 20-50 tokens each for tool names and descriptions. The update eliminated the primary argument that Skills were superior due to lower token usage.
Skills Provide Procedural Knowledge, MCP Manages External Tools
Developers reached consensus on distinct roles: Skills deliver procedural knowledge teaching AI how to perform tasks, using 30-50 tokens each loaded on-demand. MCP manages external tool connections providing live data access to current system states. As one developer explained, "Skills provide 'here's how to do things,' Projects/context provide 'here's what you need to know,' and MCP provides deterministic ways for LLMs to speak to external systems."
Platform Compatibility Favors MCP for Cross-Platform Work
A key practical distinction emerged around platform compatibility. Developers noted that "Skills that need a CLI won't even work on ChatGPT or web Claude. MCP just connects and works from anywhere." This gives MCP broader applicability across different AI platforms, while some Skills remain Claude Code-specific.
Developer Resources Document Complementary Architecture
Multiple guides published in March 2026 documented how to combine both approaches. Resources included "MCP is dead or MCP vs Skills — revisited" by Alon Nisser on Medium, "Claude Code Skills vs MCP vs Plugins: Complete Guide 2026" on MorphLLM, and "MCP vs. Agent Skills in the Era of Claude Code" by Atal Upadhyay. The official Claude blog published "Skills explained: How Skills compares to prompts, Projects, MCP, and subagents" to clarify the relationships.
Community Building Integrated Systems Using Both Technologies
Developers shared integrated architectures combining Skills, MCP, hooks, and projects. One developer reported: "In 2026, Claude became my co-founder. Not an assistant, a system running 40% of my agency ops. I packaged everything into one resource: Claude Projects Architecture, Claude Code Setup, Claude + n8n MCP, Claude Skills Blueprint, Query MCP + SEO MCPs." This post received 64 likes and 26 retweets, indicating significant community interest in integrated approaches.
Another developer explained the full stack: "wait until you discover skills, hooks, and MCP servers — that's where Claude Code goes from 'useful tool' to 'platform.' Skills are just markdown files that define workflows. Hooks are scripts that fire at lifecycle points. MCP servers connect external tools. All native, all [integrated]."
Key Takeaways
- Hacker News discussion with 319 points and 269 comments on April 10, 2026, clarified that MCP and Skills serve complementary rather than competing roles in Claude development
- MCP adopted progressive discovery in January 2026, reducing token usage from 50,000+ to 20-50 tokens per tool and eliminating the efficiency advantage Skills previously held
- Skills provide procedural knowledge (how to do things) using 30-50 tokens each, while MCP manages external tool connections providing live data access to current system states
- MCP offers broader platform compatibility working across ChatGPT and web Claude, while some CLI-dependent Skills remain Claude Code-specific
- Developers are building integrated architectures combining Skills, MCP, hooks, and projects rather than choosing one technology over another