Mcp Server skills for AI agents

Securing MCP servers with authentication, authorization, and defensive practices. Covers OAuth 2.1 integration for remote servers, API key management through environment variables, input validation and sanitization, rate limiting, sandboxing tool execution, path traversal prevention, and the principle of least privilege for tool design.

Deploying MCP servers across different environments and transports. Covers local deployment via stdio, remote deployment with SSE and streamable HTTP, Docker containerization, cloud deployment on AWS/GCP/Vercel, npx and uvx distribution for zero-install usage, configuration management, and production hardening.

Core architecture of the Model Context Protocol (MCP) — the open protocol from Anthropic that connects AI assistants to external tools and data sources. Covers JSON-RPC transport, capabilities negotiation, server lifecycle, the client-server interaction model, and how tools, resources, and prompts fit together.

Common architectural patterns for MCP servers — database servers, API wrappers, file system servers, multi-tool orchestration, caching strategies, error recovery, and composition patterns. Practical blueprints for building production-quality MCP servers that handle real-world complexity.

Defining prompt templates in MCP servers that AI clients can discover and invoke. Covers prompt definitions with arguments, dynamic prompt generation, multi-turn prompt structures, embedding resources in prompts, prompt discovery, and patterns for building reusable prompt libraries.

Building MCP servers in Python using the official mcp SDK and the FastMCP high-level pattern. Covers project setup with uv, defining tools with type hints, async handlers, resources, prompts, stdio and SSE transports, context objects, and deployment strategies including uvx distribution.

Exposing data and content to AI clients through MCP resources. Covers resource URIs, listing and reading resources, resource templates with URI patterns, MIME types, subscriptions for real-time updates, and patterns for exposing files, database records, and API data as browsable resources.

Use this skill when auditing the security of a Model Context Protocol server or client. Activate when users mention mcp server security audit, MCP vulnerabilities, tool poisoning, prompt injection via MCP, confused deputy, token passthrough, OAuth for MCP, rug-pull tool descriptions, or reviewing an MCP integration before connecting it to an agent with real capabilities.

Testing and debugging MCP servers effectively. Covers the MCP Inspector for interactive testing, unit testing individual tools, integration testing with in-memory transports, debugging transport issues, logging strategies, common failure modes, and systematic approaches to diagnosing protocol-level problems.

Defining and implementing tools in MCP servers — the primary way AI assistants take actions through MCP. Covers tool definitions with JSON Schema inputs, writing tool handlers, returning structured results, error handling with isError, tool annotations for UI hints, and patterns for robust tool implementations.

Building MCP servers in TypeScript using the official @modelcontextprotocol/sdk package. Covers project setup, the McpServer high-level API, defining tools with Zod schemas, stdio and SSE transports, streaming responses, error handling, and deployment as npm packages or standalone binaries.