MCP Integration Skill: OpenAI Agents + Model Context Protocol

Overview

This skill teaches how to build MCP (Model Context Protocol) tools for OpenAI Agents SDK, enabling AI agents to access external resources, databases, and APIs through standardized tool interfaces.

🛑 MANDATORY REFERENCE

ALWAYS check these files before creating MCP tools:

1. •CLAUDE.md: Usage guidelines and core constraints
2. •TOKENS.md: Required packages, environment setup, security patterns
3. •INTEGRATION_PATTERNS.md: Server structure, tool patterns, integration flow
4. •CODE_TEMPLATES.md: Code templates for tools, agents, auth

Core Capabilities

• •Create MCP servers with FastMCP
• •Integrate with OpenAI Agents SDK dynamically
• •Handle JWT authentication for user isolation
• •Build user-scoped tools for databases/APIs
• •Manage server lifecycle (connect/cleanup)
• •Error handling and testing patterns

When to Use This Skill

User requests:

• •"Create an MCP tool for [resource]"
• •"Add [database/API] access to my agent"
• •"Build a tool that can [action]"
• •"How do I authenticate MCP with JWT?"
• •"Connect my agent to [external service]"

Common Patterns & Responses

User: "Create an MCP tool to manage user files" You: Generates FastMCP server with file tools, integrates with agent, adds JWT auth, provides test suite

User: "My agent needs database access" You: Creates SQLModel tools, user-scoped queries, connection pooling, error handling

User: "How do I add authentication to MCP tools?" You: Shows JWT validation pattern, user_id extraction, secure context injection