Creating Your First MCP Server

This is the deck presented at the 1st session of the MS Reactor MCP Bootcamp: https://developer.microsoft.com/en-us/reactor/events/26169/

Transcript

1. Creating Your First MCP Server MCP Bootcamp https://aka.ms/mcpbootcamp Carlotta Castelluccio – Senior AI Advocate @ Microsoft
2. Agenda 1. Why another protocol? 2. Model Context Protocol (MCP): Deep Dive 3. Building your MCP server 4. AI Agents and MCP 5. MCP at Microsoft
3. Why another protocol?
4. MCP is an open protocol created by Anthropic that standardizes how applications provide context to LLMs. It’s like USB-C but for AI
5. The need for MCP arises – a Timeline Integration chaos!!! Tools are integral to LLM performance The need for a standard (MCP)
6. Parts of MCP Clients Servers Hosts
7. But how do we provide real time context? How do we connect to our database? Query GitHub Issues? Run tests with natural language? How can we bridge the gaps between our code, the client and LLM, and everything else? Agent mode Agentic coding Accelerate delivery by shifting from code to outcomes Autonomous AI pair programmer that plans, executes, and iterates on multi-step coding tasks – from editor to terminal, tests to debugger.
8. What is a Protocol? http:// http:// HTTP request GET /wiki/Mount_Kilimanjaro HTTP Response Server Client Browser
9. What is a Protocol? HTTP request GET /api/mountains/kilimanjaro HTTP Response Server REST Client { "name": "Mount Kilimanjaro", "elevation": 5895, "location": "Tanzania" }
10. MCP is the HTTP for LLMs! MCP Host MCP Client LLM MCP Server MCP Server MCP Server Tool Resource Prompt IN OUT { name: string; description?: string; inputSchema: type: "object", properties: { ... } } } { uri: string; name: string; description?: string; mimeType?: string; } { name: string; description?: string; arguments?: [ { name: string; description?: string; required?: boolean; } ] }
11. Model Context Protocol (MCP): Deep Dive
12. MCP Deep-Dive Adapted from Mahesh Murag (Anthropic) AI Engineer session MCP Client  Invokes Tools  Queries for Resources  Interpolates Prompts MCP Server  Exposes Tools  Exposes Resources  Exposes Prompts Tools Model-controlled Functions invoked by the model Resources Application-controlled Data exposed to the application Prompts User-controlled Pre-defined templates for AI interactions Retrieve / search Send a message Update DB records Files Database records API Responses Document Q&A Transcript Summary Output as JSON
13.  Functions that LLMs can call to perform specific actions - basically function calling. • Send email • Update todos • Run tests • File issue bug The popular basic primitive Tools
14. Tools: User controls Provide user controls to fine-tune and improve tool calls Per-chat tool selection Manually reduce tools required to complete a specific task Mention tools in prompts Reduce the guessing and reference tools as needed. User-defined tool sets Re-use task-specific tools across tasks
15.  Expose data in a read-only fashion as URIs • Files • Documents • Database entries/schemas • Images Ease of adding and returning data Resources
16. Resources: Use cases Make data accessible to clients and users Reduce response tokens Returning embedded resources lets the agent pull the data. Expose data/files to user Provide to assets for the user to act on, not LLM context. Attach as context Reducing tool lookups with specific resources attached.
17.  Pre-defined templates for AI tasks, explicitly invoked by the user. • Static presets • Reusable with placeholders • Dynamically generated Quick Prompts
18.  /welcome  /setup  /need-info {issue}  /impl-metric {name}  /migrate-db …  /root-cause-wizard Prompts: Use cases From static templates to dynamically generates workflows Onboarding prompts Welcome prompts for users to verify setup and tour the tools Common workflows One-shot tasks, parameterized for reusability Context-aware workflows Dynamically customized for the current user/codebase
19. MCP Flow
20. Building your MCP server
21. Official SDKs
22. Official Python SDK for MCP Clients Servers Local – Stdio Transport aka.ms/python-sdk-mcp Remote – Server Sent Events Remote – Streamable HTTP
23. Requirements from fastmcp import FastMCP pip install “mcp-cli” npx @modelcontextprotocol/inspector
25. Demo aka.ms/mcp-for-beginners 1. Building your first MCP server in Python 2. Building your first MCP client in Python and make it interact with the server
26. AI Agents and MCP
27. User experience and/or Trigger Orchestrator Knowledge Grounding and memory Tools Actions, triggers, workflows Autonomy Planning, exceptions, self- learning Foundation models Integration chaos! System Message (this is the “job description” of your agent!) MCP as AI Agent’s tools
28. User experience and/or Trigger Orchestrator Knowledge Grounding and memory Tools Actions, triggers, workflows Autonomy Planning, exceptions, self- learning Foundation models Integration chaos! System Message (this is the “job description” of your agent!) MCP as AI Agent’s tools MCP
29. MCP at Microsoft
30. Model Context Protocol At Microsoft Model Context Protocol MPC enables seamless Integration between LLM Apps and external data sources Azure API Management Build Local & Remote MCPs Tools Community Enterprise Windows MCP Clients & Host Copilot Studio VS Code Foundry Agent Visual Studio Semantic Kernel Registry GitHub Copilot C# MCP SDK Microsoft Connectors Azure Functions Microsoft Teams Windows
31. Risks and security best practices Risks: - Prompt injection & Tool Poisoning - Authentication & Tokens management - Excessive permissions - Supply Chain Security Mitigations: - Azure Content Safety & Prompt Shields - Microsoft Entra ID - RBAC & least privilege - GitHub Advanced Security, Azure DevOps
32. Today's tutorial aka.ms/mcp-for-beginners Python SDK for MCP aka.ms/python-sdk-mcp Model Context Protocol modelcontextprotocol.io Get support https://aka.ms/foundrydevs Learn more
33. https://aka.ms/mcpbootcamp