langchain4j-mcp-server-patterns

Name: langchain4j-mcp-server-patterns
Author: giuseppe-trisciuoglio

1.6kinstalls36forks

How to install langchain4j-mcp-server-patterns?

giuseppe-trisciuoglio/developer-kit/langchain4j-mcp-server-patterns

$npx -y skills add giuseppe-trisciuoglio/developer-kit --skill langchain4j-mcp-server-patterns

Installs into the current project.

Summary

Provides LangChain4j patterns for implementing MCP (Model Context Protocol) servers, creating Java AI tools, exposing tool calling capabilities, and integrating MCP clients with AI services. Use when building a Java MCP server, implementing tool calling in Java, connecting LangCh

Files · 1

View on GitHub

SKILL.md

1	# LangChain4j MCP Server Implementation Patterns
2
3	## Overview
4
5	Use this skill to design and implement Model Context Protocol (MCP) integrations with LangChain4j.
6
7	The main concerns are:
8	- defining a clean tool, resource, and prompt surface
9	- choosing the right transport and bootstrap model
10	- filtering unsafe capabilities before exposing them to agents or applications
11
12	Keep `SKILL.md` focused on the implementation flow. Use the bundled references for expanded examples and API-level detail.
13
14	## When to Use
15
16	Use this skill when:
17	- building a Java MCP server that exposes tools, resources, or prompts
18	- integrating LangChain4j with one or more external MCP servers
19	- wiring MCP support into a Spring Boot application
20	- filtering available tools by tenant, user role, or runtime context
21	- adding observability, resilience, and safe failure handling around MCP interactions
22	- reviewing an MCP integration for prompt-injection and side-effect risks
23
24	Typical trigger phrases include `langchain4j mcp`, `java mcp server`, `mcp tool provider`, `spring boot mcp`, and `connect langchain4j to mcp`.
25
26	## Instructions
27
28	### 1. Design the MCP surface before writing code
29
30	Decide what the server should expose:
31	- tools for actions with clear inputs and side effects
32	- resources for read-only or structured data access
33	- prompts only when a reusable template adds real value
34
35	Keep names stable, descriptions concrete, and schemas small enough for a client or model to understand quickly.
36
37	### 2. Implement providers with narrow responsibilities
38
39	Use separate classes for each concern:
40	- tool provider for executable functions
41	- resource provider for discoverable and readable data
42	- prompt provider for reusable prompt templates
43
44	Validate arguments before execution and return clear error messages for invalid input or unavailable dependencies.
45
46	### 3. Choose the transport intentionally
47
48	Use:
49	- stdio for local integrations, CLI tools, and sidecar processes
50	- HTTP or SSE for remote or shared services
51
52	Pin external server versions and document how the process is started, authenticated, and monitored.
53
54	### 4. Bridge MCP into LangChain4j carefully
55
56	When consuming MCP servers from LangChain4j:
57	- initialize clients during application startup
58	- cache tool lists only when stale metadata is acceptable
59	- filter tools by trust level, environment, or user permissions
60	- fail closed for dangerous tools rather than exposing everything by default
61
62	### 5. Add resilience and security controls
63
64	At minimum:
65	- bound execution time for external calls
66	- log server and tool identity for each failure
67	- sanitize content returned by external resources before using it downstream
68	- isolate privileged tools behind allowlists, qualifiers, or role checks
69
70	### 6. Validate the full workflow
71
72	Before shipping:
73	- verify tool discovery and invocation with a real MCP client
74	- test disconnected or slow server behavior
75	- confirm that tool filtering matches the intended authorization model
76	- check that prompts and resources do not leak secrets or unsafe instructions
77
78	## Examples
79
80	### Example 1: Minimal tool provider and stdio server bootstrap
81
82	```java
83	class WeatherToolProvider implements ToolProvider {
84
85	@Override
86	public List<ToolSpecification> listTools() {
87	return List.of(
88	ToolSpecification.builder()
89	.name("get_weather")
90	.description("Return the current weather for a city")
91	.inputSchema(Map.of(
92	"type", "object",
93	"properties", Map.of(
94	"city", Map.of("type", "string")
95	),
96	"required", List.of("city")
97	))
98	.build()
99	);
100	}
101
102	@Override
103	public String executeTool(String name, String arguments) {
104	return weatherService.lookup(arguments);
105	}
106	}
107
108	MCPServer server = MCPServer.builder()
109	.server(new StdioServer.Builder())
110	.addToolProvider(new WeatherToolProvider())
111	.build();
112
113	server.start();
114	```
115
116	Use this pattern for local tool execution or a sidecar process started by another application.
117
118	### Example 2: Expose MCP tools to a LangChain4j AI service with filtering
119
120	```java
121	McpToolProvider toolProvider = McpToolProvider.builder()
122	.mcpClients(mcpClients)
123	.failIfOneServerFails(false)
124	.filter((client, tool) -> !tool.name().startsWith("admin_"))
125	.build();
126
127	Assistant assistant = AiServices.builder(Assistant.class)
128	.chatModel(chatModel)
129	.toolProvider(toolProvider)
130	.build()

People who liked this also viewed

6 picks

Type

langchain4j-mcp-server-patterns

1.6kinstalls36forks

View on GitHub

How to install langchain4j-mcp-server-patterns?

giuseppe-trisciuoglio/developer-kit/langchain4j-mcp-server-patterns

$npx -y skills add giuseppe-trisciuoglio/developer-kit --skill langchain4j-mcp-server-patterns

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# LangChain4j MCP Server Implementation Patterns
2
3	## Overview
4
5	Use this skill to design and implement Model Context Protocol (MCP) integrations with LangChain4j.
6
7	The main concerns are:
8	- defining a clean tool, resource, and prompt surface
9	- choosing the right transport and bootstrap model
10	- filtering unsafe capabilities before exposing them to agents or applications
11
12	Keep `SKILL.md` focused on the implementation flow. Use the bundled references for expanded examples and API-level detail.
13
14	## When to Use
15
16	Use this skill when:
17	- building a Java MCP server that exposes tools, resources, or prompts
18	- integrating LangChain4j with one or more external MCP servers
19	- wiring MCP support into a Spring Boot application
20	- filtering available tools by tenant, user role, or runtime context
21	- adding observability, resilience, and safe failure handling around MCP interactions
22	- reviewing an MCP integration for prompt-injection and side-effect risks
23
24	Typical trigger phrases include `langchain4j mcp`, `java mcp server`, `mcp tool provider`, `spring boot mcp`, and `connect langchain4j to mcp`.
25
26	## Instructions
27
28	### 1. Design the MCP surface before writing code
29
30	Decide what the server should expose:
31	- tools for actions with clear inputs and side effects
32	- resources for read-only or structured data access
33	- prompts only when a reusable template adds real value
34
35	Keep names stable, descriptions concrete, and schemas small enough for a client or model to understand quickly.
36
37	### 2. Implement providers with narrow responsibilities
38
39	Use separate classes for each concern:
40	- tool provider for executable functions
41	- resource provider for discoverable and readable data
42	- prompt provider for reusable prompt templates
43
44	Validate arguments before execution and return clear error messages for invalid input or unavailable dependencies.
45
46	### 3. Choose the transport intentionally
47
48	Use:
49	- stdio for local integrations, CLI tools, and sidecar processes
50	- HTTP or SSE for remote or shared services
51
52	Pin external server versions and document how the process is started, authenticated, and monitored.
53
54	### 4. Bridge MCP into LangChain4j carefully
55
56	When consuming MCP servers from LangChain4j:
57	- initialize clients during application startup
58	- cache tool lists only when stale metadata is acceptable
59	- filter tools by trust level, environment, or user permissions
60	- fail closed for dangerous tools rather than exposing everything by default
61
62	### 5. Add resilience and security controls
63
64	At minimum:
65	- bound execution time for external calls
66	- log server and tool identity for each failure
67	- sanitize content returned by external resources before using it downstream
68	- isolate privileged tools behind allowlists, qualifiers, or role checks
69
70	### 6. Validate the full workflow
71
72	Before shipping:
73	- verify tool discovery and invocation with a real MCP client
74	- test disconnected or slow server behavior
75	- confirm that tool filtering matches the intended authorization model
76	- check that prompts and resources do not leak secrets or unsafe instructions
77
78	## Examples
79
80	### Example 1: Minimal tool provider and stdio server bootstrap
81
82	```java
83	class WeatherToolProvider implements ToolProvider {
84
85	@Override
86	public List<ToolSpecification> listTools() {
87	return List.of(
88	ToolSpecification.builder()
89	.name("get_weather")
90	.description("Return the current weather for a city")
91	.inputSchema(Map.of(
92	"type", "object",
93	"properties", Map.of(
94	"city", Map.of("type", "string")
95	),
96	"required", List.of("city")
97	))
98	.build()
99	);
100	}
101
102	@Override
103	public String executeTool(String name, String arguments) {
104	return weatherService.lookup(arguments);
105	}
106	}
107
108	MCPServer server = MCPServer.builder()
109	.server(new StdioServer.Builder())
110	.addToolProvider(new WeatherToolProvider())
111	.build();
112
113	server.start();
114	```
115
116	Use this pattern for local tool execution or a sidecar process started by another application.
117
118	### Example 2: Expose MCP tools to a LangChain4j AI service with filtering
119
120	```java
121	McpToolProvider toolProvider = McpToolProvider.builder()
122	.mcpClients(mcpClients)
123	.failIfOneServerFails(false)
124	.filter((client, tool) -> !tool.name().startsWith("admin_"))
125	.build();
126
127	Assistant assistant = AiServices.builder(Assistant.class)
128	.chatModel(chatModel)
129	.toolProvider(toolProvider)
130	.build()

People who liked this also viewed

6 picks

Type