🛠️ Skill: mcp-expert (v1.4.0)

Executive Summary

mcp-expert is the definitive resource for building and managing Model Context Protocol (MCP) ecosystems. In 2026, MCP is the backbone of autonomous AI agents, enabling them to navigate complex data, interact with legacy APIs, and render dynamic user interfaces. This skill covers the entire lifecycle—from proactive onboarding to advanced MCP App development and high-security server orchestration.

🚀 Core Capabilities

Proactive Onboarding: Automatically detecting and configuring missing MCP servers.
Agentic Tool Design: Crafting outcome-oriented tool interfaces for LLM excellence.
Multimodal Interaction: Leveraging resources for long-form data (Video, PDF) via MCP URIs.
Interactive UI Rendering: Implementing MCP Apps for direct user-agent interaction.
Resilient Orchestration: Managing tool timeouts, pagination, and error recovery.

🚫 The "Do Not" List (Anti-Patterns)

Anti-Pattern	Why it fails in 2026	Modern Alternative
Chatty APIs	Wastes LLM tokens and increases latency.	Use Outcome-Oriented Tools.
Raw Exceptions	LLM cannot self-correct from stack traces.	Return Helpful Error Strings.
Hardcoded Secrets	Massive security vulnerability.	Use Environment Variable Mapping.
Stdout Logging	Breaks the Stdio transport channel.	Log only to Stderr.
Monolithic Servers	Hard to maintain and discover.	Use Focused, Small Servers (5-15 tools).

⚡ Quick Start: Onboarding Protocol

When an agent needs a new capability (e.g., "Search the web"):

Validation: Check if the required MCP server is already active.
Guide: If missing, provide the user with a direct installation path.
- "I need to browse the web. I'll setup the browser-use MCP server now."
Config: Use uvx for zero-install execution where possible.

"browser-use": {
  "command": "uvx",
  "args": ["mcp-server-browser-use", "server"],
  "env": { "BROWSER_USE_API_KEY": "..." }
}

🛠 Standard Production Patterns

Pattern A: The "Progressive Disclosure" Resource

For large datasets, don't return all data. Return a list of URIs.

mcp://sales/2026/summary
mcp://sales/2026/detailed-report

Pattern B: The "Retryable" Tool

Tools that return a retry_with suggestion if inputs are slightly off.

⚙️ Professional Server Engineering

Building a high-performance server in 2026:

Runtime: Bun.
Validation: Zod.
Standard: 2025-11-25 Specification.

See References: Server Development for the blueprint.

🔒 Security & Zero-Trust Auth

OAuth 2.1: Mandatory for enterprise data access.
Capability Scopes: Limit the agent to the task at hand.
HITL Gate: Human-in-the-loop approval for destructive actions.

See References: Security & Auth for implementation.

🖥️ MCP Apps: Interactive Outcomes

MCP Apps are the next evolution of AI interfaces.

Live Dashboards.
Interactive File Editors.
Visual Selection Tools.

See References: MCP Apps for architectural details.

📖 Reference Library

Detailed deep-dives into MCP excellence:

Server Development: Professional engineering standards.
Security & Auth: Protecting your data and tools.
MCP Apps: Interactive UI standard for 2026.
Troubleshooting Guide: Debugging complex interactions.

Updated: January 22, 2026 - 17:35

mcp-expert

Safety Notice

Copy this and send it to your AI assistant to learn