LangChain4j

Overview

LangChain4j is an idiomatic Java/JVM library for building LLM-powered applications. It provides unified APIs over model providers and embedding stores, plus higher-level abstractions for chat memory, tools, RAG, AI services, agentic workflows, MCP tool providers, and framework integrations.

This is a high-value directory gap because most agent framework coverage is Python or TypeScript-heavy. LangChain4j targets Java teams that want typed APIs, POJOs, annotations, interfaces, dependency injection, Maven/Gradle modules, and enterprise framework support without porting Python agent examples into a JVM codebase.

Install

LangChain4j is distributed as Java modules. Add the module coordinates needed for the selected model, tool, RAG, MCP, or framework integration.

Base Maven shape:

<dependency>
    <groupId>dev.langchain4j</groupId>
    <artifactId>langchain4j</artifactId>
</dependency>

MCP Docker stdio support uses a separate module:

<dependency>
    <groupId>dev.langchain4j</groupId>
    <artifactId>langchain4j-mcp-docker</artifactId>
</dependency>

Check the official docs and Maven Central for current versions and module names, especially when mixing core, beta, community, Spring Boot, Quarkus, Helidon, or Micronaut dependencies.

Agent Capabilities

MCP Fit

LangChain4j is not a standalone MCP server entry. It belongs in tools because it lets JVM applications consume MCP tools and connect them to AI services. The MCP tutorial documents McpTransport, McpClient, McpToolProvider, tool-name filters, tool-name mapping, multiple MCP clients, and dynamic client/filter management.

That makes it useful for Java teams building Claude-adjacent agent products where MCP servers provide the external tool surface. It also means teams need strict permission design: use read-only tools first, filter tools by name, map conflicting tool names, and avoid handing broad filesystem, Docker, SaaS, or database access to a model loop without review.

Use Cases

• Build Java or Kotlin LLM applications with model-provider abstraction.
• Add RAG over internal docs, support content, code search, or product data.
• Connect Spring Boot or Quarkus services to model calls and tool calling.
• Let AI services call selected MCP tools through stdio, HTTP, WebSocket, or Docker stdio transports.
• Compose typed agentic workflows with shared state and subagents.
• Use enterprise JVM observability, tests, dependency management, and deployment practices around agent workflows.

Source Review

Verified on 2026-06-18:

• GitHub reports langchain4j/langchain4j as an Apache-2.0 repository with active development, 12,000+ stars, and latest release tag 1.16.3.
• The repository description and README identify LangChain4j as an idiomatic Java/JVM library for LLM applications with unified model APIs, embedding stores, tool calling, agents, RAG, and enterprise Java integrations.
• The README says LangChain4j supports 20+ LLM providers, 30+ embedding stores, low-level prompt/chat-memory/function-calling abstractions, high-level agents and RAG patterns, and examples for plain Java, Quarkus, Spring Boot, Helidon, and Micronaut.
• The agents tutorial documents the experimental langchain4j-agentic module, @Agent, AgenticServices, AgenticScope, sequential workflows, loop workflows, and shared variables across agents.
• The MCP tutorial documents stdio, Streamable HTTP, WebSocket, Docker stdio, and legacy HTTP/SSE transports, plus McpClient, McpToolProvider, tool-name filtering, tool-name mapping, and binding MCP tools into AI services.
• The RAG tutorial documents indexing, retrieval, Easy RAG, naive RAG, advanced RAG, embedding stores, document loading, splitting, and retrieval patterns.

Safety and Privacy

LangChain4j runs inside the JVM application boundary, so its safety depends on normal application engineering: dependency pinning, secrets management, tool authorization, logging policy, tenant isolation, and tests. Treat model-visible tools as public method surfaces until each one has scoped input validation, authorization, rate limiting, and audit logging.

MCP integration deserves extra review. A stdio server may start a local subprocess, a Docker transport may touch container runtime access, and HTTP or WebSocket transports may send user context to remote services. Filter tools aggressively and avoid exposing write-capable tool sets before an operator has a clear approval path.

Duplicate Check

Checked current content/tools/, content/mcp/, content/agents/, content/skills/, guides, collections, open pull requests, and repository-wide content for langchain4j/langchain4j, LangChain4j, dev.langchain4j, langchain4j-agentic, Java AI agents, Java RAG framework, Spring Boot AI agents, Quarkus LangChain4j, LangChain4j MCP, and JVM LLM framework. Existing entries cover adjacent Python/TypeScript frameworks, RAG stores, MCP SDKs, and agent platforms such as LangChain, LangGraph, LlamaIndex, Haystack, Agno, Pydantic AI, Mastra, CrewAI, AutoGen, CAMEL, Smolagents, and the official MCP SDKs, but no dedicated LangChain4j tools entry, exact source URL duplicate, target file, or open duplicate PR was found.