Skills, Tools and MCPs - What’s The Difference?

2023-09-06 · Source: Artificial Ignorance · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Emerging Technologies & Innovation · Depth: Intermediate, medium

Summary

OpenAI's function calling for GPT-4, released two and a half years ago, enabled language models to execute real-world actions via API calls and data manipulation. This foundational capability led to features like ChatGPT Plugins, Code Interpreter, and web browsing. However, initial implementations faced scalability issues due to bespoke integrations and single-function call limitations. The Model Context Protocol (MCP), proposed by Anthropic in late 2024, emerged as a standardized solution, offering dynamic discovery, richer primitives like streaming and persistent context, and event-driven updates. While MCP addressed interoperability and distribution, it introduced security and "judgment" challenges. Subsequently, Skills (or Agent Skills) were developed to provide models with learned expertise and guidance on effectively using capabilities, often through structured markdown files like `SKILL.md` that include detailed instructions and example workflows. This evolution from basic tools to standardized infrastructure (MCP) and knowledge layers (Skills) represents a progression towards building more sophisticated, AI-native products.

Key takeaway

For AI/ML Directors building complex agentic systems, understanding the distinct roles of Tools, MCPs, and Skills is crucial. Your team should architect solutions by separating atomic capabilities (Tools), standardizing their access and interaction (MCP), and explicitly encoding operational expertise (Skills) to move beyond bespoke prompting and build scalable, AI-native applications. This layered approach mitigates the challenges of integration complexity and poor tool utilization.

Key insights

AI agent capabilities evolve from basic function calls to standardized protocols and explicit expertise layers.

Principles

• Capabilities are easier to give models than to manage.
• Standardization improves distribution, not quality.
• Models need guidance on effective tool usage.

Method

Tools provide atomic capabilities, MCP offers standardized infrastructure for access and discovery, and Skills supply the knowledge layer for effective tool utilization and domain expertise.

In practice

• Use JSON schema to describe function interfaces.
• Implement MCP servers for dynamic tool discovery.
• Package expertise into `SKILL.md` files for models.

Topics

• Function Calling
• Model Context Protocol
• AI Skills
• AI Agents
• API Integration

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, Machine Learning Engineer, AI Architect

Related on AIssential

• MCP vs Agent Skills: Different Altogether — MCP and Agent Skills are complementary, not competing, technologies for AI agent development.
• AI Agents, Tools, MCP, and Skills: The Core, The Embellishment, and The Gimmick — AI agent development requires distinguishing core components like Agents and Tools from less practical elements like MCP.
• Why MCP is dead & How I vibe now — Skill-based CRI tools drastically cut token consumption for coding agents, enhancing scalability and efficiency.
• MCP vs ADK: How Modern AI Agents Connect and Work Together — MCP standardizes LLM connectivity to external tools, while ADK structures agent building and orchestration.
• Tools, Resources and Prompts — MCP defines Tools, Resources, and Prompts as core capabilities for structured AI interaction and control.
• From Silos to Symphony: The Model Context Protocol and the Dawn of the Unified Machine — The Model Context Protocol (MCP) unifies AI model interaction with external tools and data through an open, standardized interface.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Ignorance.