langchain4j-mcp-server-patterns

The langchain4j-mcp-server-patterns skill provides patterns for designing and implementing Model Context Protocol (MCP) integrations with LangChain4j in Java, solving the problem of exposing tools, resources, and prompts to AI agents cleanly and safely. It covers building Java MCP servers, connecting LangChain4j to external MCP servers, wiring MCP into Spring Boot applications, and reviewing integrations for prompt-injection and side-effect risks.

The skill follows an implementation flow: designing the MCP surface (tools for actions, resources for read-only data, prompts for reusable templates) before writing code; implementing providers with narrow responsibilities and validated arguments; choosing transports intentionally (stdio for local, HTTP or SSE for remote) and pinning external server versions; bridging MCP into LangChain4j by initializing clients at startup and filtering tools by trust level, environment, or user permissions; and adding resilience and security controls. Notably, the security guidance is defensive: it recommends failing closed for dangerous tools, bounding execution time, sanitizing content returned by external resources before downstream use, isolating privileged tools behind allowlists and role checks, and confirming that prompts and resources do not leak secrets. It ships a SKILL.md, an MCP server Java template, an API reference, and examples, and declares safe allowed-tools including WebFetch.

It targets Java and Spring Boot developers building agent tooling who need a secure, well-structured MCP surface. Use cases include tenant-scoped tool exposure, resilient external-server integration, and security reviews of existing MCP integrations. The emphasis on safe defaults and prompt-injection mitigation makes this defensive rather than offensive tooling.