Bonus lab 24: MCP

Author: siddhi

bruno/mcp/1_initialize.bru

@@ -0,0 +1,37 @@
+meta {
+  name: 1. Initialize Session
+  type: http
+  seq: 1
+}
+
+post {
+  url: http://localhost:9000/mcp
+  body: json
+  auth: none
+}
+
+headers {
+  Accept: application/json, text/event-stream
+  Content-Type: application/json
+}
+
+body:json {
+  {
+    "jsonrpc": "2.0",
+    "id": "init-001",
+    "method": "initialize",
+    "params": {
+      "protocolVersion": "2024-11-05",
+      "capabilities": {
+        "roots": {
+          "listChanged": false
+        },
+        "sampling": {}
+      },
+      "clientInfo": {
+        "name": "mcp-client",
+        "version": "1.0.0"
+      }
+    }
+  }
+}

bruno/mcp/2_initialized.bru

@@ -0,0 +1,24 @@
+meta {
+  name: 2. Send Initialized Notification
+  type: http
+  seq: 2
+}
+
+post {
+  url: http://localhost:9000/mcp
+  body: json
+  auth: none
+}
+
+headers {
+  Accept: application/json, text/event-stream
+  Content-Type: application/json
+  mcp-session-id: 
+}
+
+body:json {
+  {
+    "jsonrpc": "2.0",
+    "method": "notifications/initialized"
+  }
+}

bruno/mcp/3_list_tools.bru

@@ -0,0 +1,25 @@
+meta {
+  name: 3. List Tools
+  type: http
+  seq: 3
+}
+
+post {
+  url: http://localhost:9000/mcp
+  body: json
+  auth: none
+}
+
+headers {
+  Accept: application/json, text/event-stream
+  Content-Type: application/json
+  mcp-session-id: 
+}
+
+body:json {
+  {
+    "jsonrpc": "2.0",
+    "id": "list-001",
+    "method": "tools/list"
+  }
+}

bruno/mcp/4_call_add.bru

@@ -0,0 +1,32 @@
+meta {
+  name: 4. Call Add Tool
+  type: http
+  seq: 4
+}
+
+post {
+  url: http://localhost:9000/mcp
+  body: json
+  auth: none
+}
+
+headers {
+  Accept: application/json, text/event-stream
+  Content-Type: application/json
+  mcp-session-id: 
+}
+
+body:json {
+  {
+    "jsonrpc": "2.0",
+    "id": "add-001",
+    "method": "tools/call",
+    "params": {
+      "name": "add",
+      "arguments": {
+        "a": 5,
+        "b": 3
+      }
+    }
+  }
+}

bruno/mcp/5_call_mul.bru

@@ -0,0 +1,32 @@
+meta {
+  name: 5. Call Multiply Tool
+  type: http
+  seq: 5
+}
+
+post {
+  url: http://localhost:9000/mcp
+  body: json
+  auth: none
+}
+
+headers {
+  Accept: application/json, text/event-stream
+  Content-Type: application/json
+  mcp-session-id: 
+}
+
+body:json {
+  {
+    "jsonrpc": "2.0",
+    "id": "mul-001",
+    "method": "tools/call",
+    "params": {
+      "name": "mul",
+      "arguments": {
+        "a": 4,
+        "b": 7
+      }
+    }
+  }
+}

bruno/mcp/bruno.json

@@ -0,0 +1,9 @@
+{
+  "version": "1",
+  "name": "MCP Client",
+  "type": "collection",
+  "ignore": [
+    "node_modules",
+    ".git"
+  ]
+}

labs/24-mcp.md

@@ -0,0 +1,53 @@
+# Lab 24: Bonus: Model Context Protocol
+
+In this lab, we will create an MCP server and connect to it using Bruno
+
+## High level overview
+
+1. Install `fastmcp` package
+1. Create a file `mcp_server.py`
+1. Create an `add` tool as follows
+
+```python
+@mcp.tool()
+def add(a: int, b: int) -> int:
+    """Add two integers together.
+    
+    Args:
+        a: First integer
+        b: Second integer
+        
+    Returns:
+        The sum of a and b
+    """
+    return a + b
+```
+
+1. Now similarly create another tool of your choice. 
+  - Make sure you put all the types and use descriptive names. 
+  - Don't forget to add the documentation string. 
+  - The decorator will read all these values and provide it to the client
+
+1. Add code to run the server
+
+```python
+if __name__ == "__main__":
+    mcp.run(transport="streamable-http", host="127.0.0.1", port=9000)
+```
+
+1. Now run the server in the terminal: ```python mcp_server.py```
+
+Next, we will use Bruno to connect to this server
+
+1. Load the collection which is in the directory `bruno/mcp` into Bruno
+1. Make a call to the first API. This API connects to the server and initialises the session
+1. The initialisation will return an MCP Session ID in the response headers. Copy it
+1. Now call the second API. In the headers, fill in the session ID that you copied from the initialisation
+1. Similarly call all the remaining APIs. Understand the input and output formats
+
+These API calls happen in our agent code as well. When we configure the MCP server, openai agents sdk will
+
+- Initialise the session
+- Get a list of tools
+- These tools will be added to the prompt automatically
+- When the LLM returns a tool call, the agent will make a call to the MCP server with the defined parameters to get the response