Postman + TanStack = Amazing MCP!

Postman for Model Context Protocol (MCP) - Detailed Summary

Key Concepts:

• Model Context Protocol (MCP): A protocol for connecting language models to external tools and data sources.
• Postman: An API platform used for building and testing APIs, now also capable of handling MCP workflows.
• Tanstack Start: A full-stack framework used in the example application.
• MCP Server: A server that handles MCP requests and routes them to appropriate tools.
• MCP Tool: A function or service that performs a specific task within an MCP workflow.
• Zod: A TypeScript schema declaration and validation library.
• AI Agent: An integration within Postman that allows direct connection to AI models.
• Flow Module: A reusable component within Postman that encapsulates a sequence of actions, such as making API requests.
• Enrock: A tool that creates secure tunnels to expose local servers to the internet.

1. Introduction to Postman for MCP

• Postman is presented as a powerful tool not only for API development but also for working with Model Context Protocol (MCP).
• Two key capabilities are highlighted:
  • Debugging MCP servers.
  • Building MCP servers within Postman to connect to non-MCP APIs.

2. Building an MCP Endpoint in a Tanstack Start Application

• Application Overview: A Tanstack Start application is used as an example, which allows users to track songs they want to learn on guitar. The application persists data, but the goal is to add an MCP endpoint.
• Creating the MCP Endpoint:
  1. A new file MCP.ts is created in the routes folder.
  2. An MCP server is created using the handleMCPRequest function, which connects the MCP SDK to Tanstack Start.
  3. The route is configured to handle POST requests and forward them to the handleMCPRequest function.
• Adding Tools to the MCP Server:
  1. Helper functions addSong and getSong are imported.
  2. A tool is registered to get the list of songs.
• Testing the MCP Endpoint with Postman:
  1. The application is running on localhost:3000.
  2. In Postman, a new MCP request is created, specifying HTTP protocol and the localhost URL.
  3. The MCP Connect button is used to connect to the MCP server.
  4. The getSong tool is available and returns the list of songs.
• Adding the "Add Song" Tool:
  1. Zod (z) is imported to define the schema for the input arguments (title and artist).
  2. The addSong tool is defined to take a title and artist as input.
  3. The addSong function is called with the provided arguments, and the response is sent back.
  4. After reconnecting in Postman, the addSong tool becomes available.
• Demonstration:
  1. The "Smells Like Teen Spirit" song by Nirvana is added using the addSong tool in Postman.
  2. The song is dynamically added to the list in the browser due to an event stream between the server and the client.

3. Creating an MCP Server in Postman to Connect to an API

• Scenario: The goal is to create an MCP server in Postman that can use an existing API (e.g., /api/songs) and call out to an AI model.
• Creating a Tool Definition Scenario:
  1. A new action is created in Postman.
  2. A new scenario called "tool definition" is created.
  3. The tool definition is specified in the body of the scenario.
  4. A single tool, get song recommendations, is defined, which takes a prompt as input.
  5. The prompt is used to templatize a string (e.g., "I'd like songs about [prompt]").
• Deploying the MCP Server:
  1. The scenario is saved and deployed as "get songs recommendation".
  2. The deployment is described as fast and easy.
• Testing the MCP Server:
  1. An MCP request is created in Postman and connected to the deployed server.
  2. The get song recommendations tool is available.
  3. A prompt (e.g., "something good at a barbecue") is provided, and the templated string is returned.

4. Integrating AI with the MCP Server

• Replacing the Template with AI:
  1. The template in the tool definition is replaced with an AI agent.
  2. An AI model is selected.
  3. The AI is prompted to recommend songs.
  4. The prompt argument is connected to the AI agent, and the output is connected to the response.
• Deployment and Testing:
  1. The updated scenario is deployed.
  2. A prompt (e.g., "something good at a barbecue") is provided, and an AI-generated response is returned.
  3. The server updates live without needing to disconnect and reconnect.

5. Connecting the MCP Server to an External API

• Using Enrock to Expose the Local API:
  1. Enrock is used to create a secure tunnel to expose the local API (localhost:3000) to the internet.
  2. Enrock provides an external URL that forwards to the local server.
• Creating a Flow Module to Access the API:
  1. A flow module is created, which acts as a reusable library.
  2. A new request called "songs request" is defined within the flow module.
  3. The request is configured to call the Enrock URL for the /api/songs endpoint.
  4. The flow module extracts the body of the response as the output.
  5. The flow module is renamed to "get my songs tool".
  6. A snapshot of the flow module is created to make it available as a tool in the MCP server.
• Integrating the Flow Module into the MCP Server:
  1. The "get my songs tool" is added as a tool to the "songs recommendations" MCP server.
  2. The prompt is updated to instruct the AI to use the "get my songs tool" to get the current list of songs.
• Deployment and Testing:
  1. The updated scenario is deployed.
  2. A prompt (e.g., "something by Nirvana") is provided.
  3. The MCP server calls out to OpenAI, which in turn makes a tool request to get the songs.
  4. The flow module handles the tool request by calling the local API.
  5. The API returns the list of songs, which is sent back to OpenAI.
  6. OpenAI uses the list of songs and the prompt to generate song recommendations.

6. Conclusion

• Postman provides a powerful platform for building and testing MCP workflows.
• It allows connecting to existing APIs and integrating with AI models.
• Flow modules enable the creation of reusable components for accessing data.
• The combination of Postman, MCP, and AI enables the creation of complex and intelligent applications with minimal coding.
• Postman is a fantastic way to "no code PC" a full MCP AI integration experience.