Upgrade to Pro — share decks privately, control downloads, hide ads and more …

MCP in action ParisJUG

Zineb Bendhiba
May 13, 2025
160

MCP in action ParisJUG

Zineb Bendhiba

May 13, 2025

Transcript

35 slides
  1. MCP en action: Connecter les IA aux systèmes d’entreprise Zineb

    Bendhiba, Principal Software Engineer, Red Hat
  2. Model Context Protocol (MCP) A new protocol - A standard

    for connecting AI agents to systems - Pushed by Anthropic (Claude) USB for AI integrations
  3. Capabilities Tools - The client can invoke “tool” and get

    the response - Close to function calling, but the invocation is requested by the client - Can be anything: database, remote service… Resources - Expose data - URL -> Content Prompts - Pre-written prompt template - Allows executing specific prompt
  4. Transport JSON-RPC 2.0 - Everything is JSON - Request /

    Response and Notifications - Possible multiplexing Transports - stdio -> The client instantiates the server, sends the requests on stdio and gets the response from the same channel - Server-Sent Event (SSE) -> The client sends a POST request to the server, the response is an SSE (chunked response) - Streamable HTTP ->
  5. Quarkus MCP server extensions 2 extensions - SSE (SSE) ->

    quarkus-mcp-server-sse - stdio -> quarkus-mcp-server-stdio
  6. 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
  7. Model and Model Serving Model Model Serving Model Serving -

    Run the model - CPU / GPU - Expose an API Input - Prompt (text) - Instructions to give to the model - Taming a model is hard GPUs Input (Prompt) Output
  8. Application Model AI-infused application |ˌeɪˌaɪ ˈɪnˌfjuːzd ˌæplɪˈkeɪʃən| noun (Plural AI-Infused

    applications) A software program enhanced with artificial intelligence capabilities, utilizing AI models to implement intelligent features and functionalities. Model Serving
  9. Model <= HTTP Client <= Application HTTP Client Leaking HTTP

    stuff or abstractions - Often leak HTTP details - Error code, connection pooling - Requires implementing memory and token counting logic Application
  10. LangChain / LangChain4j / Quarkus LangChain4j LangChain LangChain4j Quarkus LangChain4j

    Inspired By Uses and extends Framework Clean separation, Managed interactions Integration Application
  11. 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
  12. Agent and MCP servers MCP servers offer new tools and

    resources that the model can call: Prompt (Context) Extend the context with MCP tools descriptions Invoke the model The model invoke a tool invocation (name + parameters) The tool is invoked via JSON-RPC and the result sent to the model The model computes the response using the tool result Response
  13. 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
  14. My obsertavtions - Most of servers are stdio - Entreprise

    vs Local - Lack of documentation for SSE - Security - Authentication - Wanaku experience : limitation in the number of tools - MCP and Agentic -> emerging A2A - Not the only protocol (ACP …)

More Decks by Zineb Bendhiba