MCP in action - Devoxx

Transcript

1. MCP in action Connecting AI to Enterprise Systems Zineb Bendhiba, IBM
2. Who am I ? Zineb Bendhiba • Apache Camel committer and PMC member • Camel Quarkus maintainer • Camel AI & Wanaku MCP router • Based in Paris - France • Twitter : @ZinebBendhiba • Bluesky : @zinebbendhiba.com • https://zinebbendhiba.com/
3. Building AI intelligent systems
4. The foundation : prompting LLM Input (prompt) Output (response) AI infused application User Process Serving APIs • Core interaction is Instruction Data response
5. The foundation : prompting LLM Input (prompt) Output (response) AI infused application User Process Serving APIs • Core interaction is Instruction Data response • Integration is stateless REST call • SDKs and librairies add value : context and history
6. The foundation : prompting LLM Input (prompt) Output (response) AI infused application User Process Serving APIs • Core interaction is Instruction Data response • Integration is stateless REST call • SDKs and librairies add value : context and history • Suitable for content generation or Q&A
7. The foundation : prompting LLM Input (prompt) Output (response) AI infused application User Process Serving APIs • Core interaction is Instruction Data response • Integration is stateless REST call • SDKs and librairies add value : context and history • Suitable for content generation or Q&A Limitation : • Cannot be an Autonomous Agent for Enterprise systems X Enterprise systems
8. How do we move from a smart Q&A service to an autonomous agent ?
9. Bridging the Gap: AI Agents with Functions LLM Input Output User Process Serving APIs • AI Agent includes predefined Functions (Tools) • Functions link Agent to Enterprise Systems Enterprise systems AI Agent List of functions
10. Bridging the Gap: AI Agents with Functions LLM Input Output User Process Serving APIs • AI Agent includes predefined Functions (Tools) • Functions link Agent to Enterprise Systems • Input: Agent sends Instruction, Context + Function Descriptions to LLM • Output : Message or Function call Enterprise systems AI Agent List of functions
11. Bridging the Gap: AI Agents with Functions LLM Input Output User Process Serving APIs • AI Agent includes predefined Functions (Tools) • Functions link Agent to Enterprise Systems • Input: Agent sends Instruction, Context + Function Descriptions to LLM • Output : Message or Function call • LLM Role: Reasoning & Decision • Creates Autonomous System Enterprise systems AI Agent List of functions
12. Bridging the Gap: AI Agents with Functions LLM Input Output User Process Serving APIs Limitations : • Functions managed inside the Agent app • Functions cannot be shared by multiple Agents • Blocks 3rd parties from exposing business logic as functions Enterprise systems AI Agent List of functions
13. How do we move from this siloed approach to a unified, enterprise wide protocol for AI interaction?
14. Model Context Protocol
15. MCP Standard for AI • Model Context Protocol • A standard for connecting AI agents to systems • Pushed by Anthropic (Claude) for AI integrations • Acts like a USB for AI integrations • Facilitates powerful connections to various systems
16. MCP : Model Context Protocol LLM User Process • Decoupling AI agent logic from the Enterprise System's logic • MCP server as universal adapter to an Enterprise system • Standardized flow • AI agent can access any capability exposed by an MCP server Enterprise systems AI Agent MCP clients MCP Server MCP Server MCP Server MCP Server
17. MCP : technical details
18. MCP Capabilities Tools • Allows the AI agents to invoke external functions • Example of tools : a database query, a REST API call, legacy service …. Resources • Expose data : content and URL Prompts • Pre-written prompt template • Allows executing specific, reusable prompt
19. JSON RPC 2.0 • Everything is JSON • Request / Response and Notifications • Language agnostic • Human readable { "jsonrpc": "2.0", "id": 2, "result": { "isError": false, "content": [ { "text": "The weather in Paris is cloudy ☁ with a temperature of 0°C. ", "type": "text" } ] } } Example of MCP server Tool Response
20. Transports: stdio • The client instantiates the server, and communications occurs over standard input/output channels • Ideal for local deployments
21. Transports : HTTP Remote communications • Server-Sent Event (SSE) ( deprecated ) -> The client sends a POST request to the server, the response is an SSE • Streamable HTTP - SSE + multiplexing
22. Create my Quarkus MCP server
23. Quarkus MCP server extensions 2 extensions - SSE (SSE / Streamable HTTP) -> quarkus-mcp-server-sse - stdio -> quarkus-mcp-server-stdio
24. Quarkus MCP server extensions @Inject CodeService codeService; @Tool(description = "Converts the string value to lower case" ) String toLowerCase(String value) { return value.toLowerCase(); } @Prompt(name = "code_assist") PromptMessage codeAssist(@PromptArg(name = "lang") String language) { return PromptMessage.withUserRole(new TextContent(codeService.assist(language))); } @Resource(uri = "file:///project/alpha" ) BlobResourceContents alpha(RequestUri uri) throws IOException{ return BlobResourceContents.create(uri.value(), Files.readAllBytes(Path.of("alpha.txt"))); } MCP Tool MCP Prompt MCP Resource
25. Demo MCP server
26. Quarkus MCP server demo Tool Get current time MCP Server Get Weather Generate a random fact Generate a greeting prompt Capabilities
27. What happens under the hood ?
28. MCP client / Response communication
31. Wanaku MCP server
32. Apache Camel Source Sink Action
33. Apache Camel and MCP : common ground Goal • Aim to connect to Any Enterprise system Integration philosophy : Lightweight • Camel defines routes via Code/XML/YAML • MCP defines integration via JSON RPC 2.0 Prompts • Camel is the swiss knife of enterprise integration (many connectors, patterns…) • MCP is the USB standard for AI integration
34. The opportunity: leveraging Camel for MCP • MCP requires robust, standardized server adapters to connect Agents to Enterprise systems • Apache Camel is already the Enterprise Integration Engine: • Camel provides 300+ pre-built Connectors (DBs, SaaS, APIs, etc.) • Transform every Camel Connector into an MCP Tool or Resource capability.
36. Wanaku : current architecture : WIP Agent MCP Server Other SSE MCP servers Camel capabilities GRPC bridge MCP Wanaku CLI Persistence cache Based on Quarkus MCP servers or any SSE server AMQP Kafka Fhir … Python C/C++ Go … Security
37. Create a Quarkus AI Infused app with Wanaku MCP server
38. LangChain / LangChain4j / Quarkus LangChain4j LangChain LangChain4j Quarkus LangChain4j Inspired By Uses and extends Framework Clean separation, Managed interactions Integration Application
39. Quarkus LangChain4j https://docs.quarkiverse.io/quarkus-langchain4j LangChain4j Quarkus LangChain4j Application LLMs Vector stores Embedding Models - Declarative clients - CDI integration - Observability (Otel, Prometheus) - Auditing - Resilience - Guardrails - RAG building blocks - Agentic architecture - Testability (unit, IT, evaluation) - Developer Joy
40. Quarkus Langchain4j MCP client extension quarkus.langchain4j.mcp.filesystem. transport-type =stdio quarkus.langchain4j.mcp.filesystem. command=npx,-y,@modelcontextprotocol/s erver-filesystem,/my-system-path/location quarkus.langchain4j.mcp.wanaku. transport-type =http quarkus.langchain4j.mcp.wanaku. url=http://my-url/mcp/sse/ stdio SSE
41. Adding an MCP ToolBox to an AIService @RegisterAiService public interface MyAiService { @McpToolBox String chat(@UserMessage String userMessage); } @RegisterAiService public interface MyAIGithubService { @McpToolBox("github") String useGithub(@UserMessage String userMessage); } Allow all Available MCP Tools Only allow some MCP Tools
42. Demo AI agent with Quarkus and Wanaku MCP server
43. Demo: AI infused app with Wanaku MCP server LLM User Process AI Agent MCP clients MCP Server Forward HTTP capability File provider Tool Get current time MCP Server Get Weather Generate a random fact Tools via OpenAPI spec
45. What happens under the hood ?
46. Quarkus MCP client with Wanaku MCP server
47. Demo resources and prompts
48. More
49. More - Tools are automatically selected by the LLMs - Prompts and Resources needs some coding logic - Additional features - Security - Authentication - Isolated MCP server or not - Description of tools - Hallucinations - Wanaku experience : limitation in the number of tools - Adding security per user - Adding agent to pre filter the Tools - Expose Camel Routes as MCP tools - Explore new ways of delivering Enterprise Integration - MCP and Agentic -> emerging A2A / ACP
50. Thank you
51. Links - https://docs.quarkiverse.io/quarkus-langchain4j/dev/index.html - https://docs.quarkiverse.io/quarkus-mcp-server/dev/index.html - https://www.wanaku.ai/ - https://quarkus.io/blog/secure-mcp-sse-server/ - https://www.anthropic.com/news/model-context-protocol - https://docs.quarkiverse.io/quarkus-langchain4j/dev/mcp.html -