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

apidays Munich 2025 - Let’s build, debug and te...

Avatar for apidays apidays
July 09, 2025
0

apidays Munich 2025 - Let’s build, debug and test a magic MCP server in Postman with a few clicks, Johannes Nicolai (Postman)

Let’s build, debug and test a magic MCP server in Postman with a few clicks
Johannes Nicolai, Strategic Solutions Engineer at Postman

apidays Munich 2025 - Accelerate AI Use Cases with APIs
July 2 & 3, 2025

------

Check out our conferences at https://www.apidays.global/

Do you want to sponsor or talk at one of our conferences?
https://apidays.typeform.com/to/ILJeAaV8

Learn more on APIscene, the global media made by the community for the community:
https://www.apiscene.io

Explore the API ecosystem with the API Landscape:
https://apilandscape.apiscene.io/

Avatar for apidays

apidays

July 09, 2025
Tweet

More Decks by apidays

Transcript

  1. All rights reserved by Postman Inc Johannes Nicolai - [email protected]

    Let’s build, debug and test a magic MCP server in Postman with a few clicks
  2. Agenda Function Calling /MCP - Physical Hello World Example Hands-On:

    Generate and run your own, magic MCP server with Harry Potter & Octocat API Hands-On: Connect your MCP Server to Claude or VSCode Hands-On: Debug, and test your MCP server Q & A 5 4 3 2 1
  3. Getting to know you better - Show off hands …

    Who is using Postman and has an account? Who is using Claude Desktop, GitHub Copilot or Cursor? Who has already heard of MCP and LLM Function Calling? Who has connected to an MCP server? Generated an MCP server? Wrote a custom MCP server? Can run node applications locally or docker locally? Can you access the slides, clone the repo and install deps? Search for “Fun APIs only” on https://postman.com/explore select Octodex-HP-MCP server, check description
  4. The Integration Challenge: The MN Problem • Each AI model

    needs unique connections to every service. • This complex setup slows progress and costs more to maintain. • Scaling AI is impossible without a simpler, unified solution
  5. Solution: Model Context Protocol MCP • MCP is a standard

    protocol for AI to use tools and data. • It simplifies integration reducing complexity to MN. • Scaling AI is impossible without a simpler, unified solution
  6. MCP  One common protocol for all models and tools

    M  N https://patel-world.medium.com/mcp-the-usb-c-of-ai-applications-a95dc77fcbf8
  7. MCP Momentum A Growing Ecosystem • MCP is backed by

    Anthropic, Google, Microsoft and OpenAI. • Many companies are rapidly adopting MCP. • A large open-source community supports MCP's growth
  8. MCP Architecture How It Works • MCP Hosts are AI

    apps where users interact. • MCP Clients manage secure connections to Servers. • MCP Servers expose capabilities from existing APIs.
  9. Postman as MCP Tool Example - Improve NYT-API and contribute

    changes back (Official Postman MCP Server)
  10. How to expose the Octocats over an API, generate test

    cases and performance test it … All in one Demo - Watch Video
  11. MCP Security Best Practices • For Developers: ◦ Prioritize secure

    credential handling (secrets managers, short-lived creds, RBAC ◦ Supply chain security (tool vetting, trusted scanners) ◦ Environment isolation (containers, network controls) ◦ Safe coding practices (input validation) • For Users: ◦ Use encrypted clients ◦ Enable comprehensive monitoring ◦ Stay vigilant (avoid unknown servers, watch for unusual activity).
  12. MCP Flow Typical Communication Flow • User Interaction: User submits

    a query to the Host. • Host Processing & Connection: Host identifies required external capabilities and connect to Server(s). • Capability Discovery & Invocation: Client queries Server for available tools and fetches capabilities to use. • Server Execution and Result: Server performs an action and returns results to Client. Client send results to Host.
  13. Postman API Network Worldʼs Largest Public API Hub # of

    Postman Users 40M # of Public Workspaces 150K # of Public APIs 450K
  14. AI adds more players to the game: Employees who are

    no dev/domain experts/native speakers can finally participate and developers get a boost, but new challenges arise
  15. One rule that will never change though: Garbage In →

    Garbage Out GenAI will just create larger FrankenstAIn solutions No AI GenAI
  16. Thousands of Postman teams/Git repos/Static files lead to API consumer

    confusion and building “FrankenstAInˮ solutions
  17. With a central internal Postman Team and Public API Network

    • See all “goodˮ lego bricks APIs/collections) available • Discover, Find, Adapt, Evaluate, Collaborate and share in real time with all stakeholders • See what other teams built with the API lego bricks, build & tests API much better • Avoid reinventing the wheel / one off integrations (consumers are no longer forced to become producers because they could not find a proper API • Start to unlock InnerSource benefits and save thousands of hours of confusion and inefficient collaboration