[Step 1/3] Copy code-interpreter-custom to prompt-agents folder

samples/python/prompt-agents/code-interpreter-custom/.env.sample

@@ -0,0 +1,3 @@
+AZURE_AI_PROJECT_ENDPOINT=
+AZURE_AI_CONNECTION_ID=
+AZURE_AI_MODEL_DEPLOYMENT_NAME=

samples/python/prompt-agents/code-interpreter-custom/.gitignore

@@ -0,0 +1,3 @@
+.venv/
+.env
+__pycache__/

samples/python/prompt-agents/code-interpreter-custom/README.md

@@ -0,0 +1,65 @@
+# Custom Code Interpreter with Session Pool MCP server
+
+This provides example Bicep code for setting up a Container Apps dynamic session pool
+with a custom code interpreter image, as well as Python client code demonstrating
+how to use it with a Foundry Hosted Agent.
+
+You will need the following installed to run the sample code:
+
+- The `az` CLI
+- Python3
+- A Python3 package manager like `uv` or `pip` + `venv`
+    - If you are using `pip`, make sure `ensurepip` is installed. On Debian/Ubuntu
+      systems, this would mean running `apt install python3.12-venv`.
+
+## Running code sample
+
+### Enable MCP server for dynamic sessions
+
+This is required to enable the preview feature.
+
+```console
+az feature register --namespace Microsoft.App --name SessionPoolsSupportMCP
+az provider register -n Microsoft.App
+```
+
+### Create a dynamic session pool with a code interpreter image
+
+Using the `az` CLI, deploy with the provided Bicep template file:
+
+```console
+az deployment group create \
+    --name custom-code-interpreter \
+    --subscription <your_subscription> \
+    --resource-group <your_resource_group> \
+    --template-file ./infra.bicep
+```
+
+> [!NOTE] This can take a while! Allocating the dynamic session pool
+> can take up to 1 hour, depending on the number of standby instances
+> requested.
+
+### Use the custom code interpreter in an agent
+
+Copy the [`.env.sample`](./.env.sample) file to `.env` and fill in the values with
+the output of the above deployment, which you can find in the Web Portal under the
+resource group.
+
+Finally, install Python dependencies and run the script:
+
+```console
+# Using uv
+
+uv sync
+uv run ./main.py
+
+# Using pip
+
+python3 -m venv .venv
+./.venv/bin/pip3 install -r requirements.txt
+./.venv/bin/python3 ./main.py
+```
+
+## Limitations
+
+File input/output and use of file stores are not directly supported in APIs, so you must use URLs (such as data URLs for small files and Azure Blob Service SAS URLs for large ones) to get data in and out.

samples/python/prompt-agents/code-interpreter-custom/infra.bicep

@@ -0,0 +1,231 @@
+@description('Suffix for resource names to ensure uniqueness')
+@minLength(3)
+param suffix string = uniqueString(resourceGroup().id)
+
+@description('Container Apps environment name')
+@minLength(3)
+param environmentName string = 'aca-env-${suffix}'
+
+@description('Session pool name')
+@minLength(3)
+param sessionPoolName string = 'sp-${suffix}'
+
+@description('The amount of CPU to provide to each container instance, in vCPU counts')
+@minValue(1)
+@maxValue(16)
+param cpu int = 1
+
+@description('The amount of RAM to provide to each container instance, in GiB')
+@minValue(1)
+@maxValue(16)
+param memory int = 2
+
+@description('Location of all ACA resources.')
+@allowed([
+  'eastus'
+  'swedencentral'
+  'northeurope'
+])
+param location string = 'swedencentral'
+
+@description('Use managed identity for deployment script principal')
+param useManagedIdentity bool = true
+
+@description('An image that implements the code interpreter HTTP API')
+param image string = 'mcr.microsoft.com/k8se/services/codeinterpreter:0.9.18-python3.12'
+
+@description('Model deployment name')
+param modelDeploymentName string = 'my-gpt-4o-mini'
+
+@description('Model to deploy')
+param modelName string = 'gpt-4o-mini'
+
+resource environment 'Microsoft.App/managedEnvironments@2025-10-02-preview' = {
+  name: environmentName
+  location: location
+  properties: {
+    workloadProfiles: [
+      {
+        name: 'Consumption'
+        workloadProfileType: 'Consumption'
+      }
+    ]
+  }
+}
+
+resource sessionPool 'Microsoft.App/sessionPools@2025-10-02-preview' = {
+  name: sessionPoolName
+  location: location
+  properties: {
+    environmentId: environment.id
+    poolManagementType: 'Dynamic'
+    containerType: 'CustomContainer'
+    scaleConfiguration: {
+      maxConcurrentSessions: 10
+      readySessionInstances: 5
+    }
+    dynamicPoolConfiguration: {
+      lifecycleConfiguration: {
+        cooldownPeriodInSeconds: 600
+        lifecycleType: 'Timed'
+      }
+    }
+    customContainerTemplate: {
+      containers: [
+        {
+          name: 'jupyterpython'
+          image: image
+          env: [
+            {
+              name: 'SYS_RUNTIME_SANDBOX'
+              value: 'AzureContainerApps-DynamicSessions'
+            }
+            {
+              name: 'AZURE_CODE_EXEC_ENV'
+              value: 'AzureContainerApps-DynamicSessions-Py3.12'
+            }
+            {
+              name: 'AZURECONTAINERAPPS_SESSIONS_SANDBOX_VERSION'
+              value: '7758'
+            }
+            {
+              name: 'JUPYTER_TOKEN'
+              value: 'AzureContainerApps-DynamicSessions'
+            }
+          ]
+          resources: {
+            cpu: cpu
+            memory: '${memory}Gi'
+          }
+          probes: [
+            {
+              type: 'Liveness'
+              httpGet: {
+                path: '/health'
+                port: 6000
+              }
+              failureThreshold: 4
+            }
+            {
+              type: 'Startup'
+              httpGet: {
+                path: '/health'
+                port: 6000
+              }
+              failureThreshold: 30
+              periodSeconds: 2
+            }
+          ]
+        }
+      ]
+      ingress: {
+        targetPort: 6000
+      }
+    }
+    mcpServerSettings: {
+      isMcpServerEnabled: true
+    }
+    sessionNetworkConfiguration: {
+      status: 'egressEnabled'
+    }
+  }
+}
+
+resource scriptPrincipal 'Microsoft.ManagedIdentity/userAssignedIdentities@2023-01-31' = if (useManagedIdentity){
+  name: 'deployScriptIdentity-${suffix}'
+  location: location
+}
+
+resource roleAssignment 'Microsoft.Authorization/roleAssignments@2022-04-01' = if (useManagedIdentity) {
+  name: guid(scriptPrincipal!.id, 'apps-sessionpool-contributor')
+  scope: resourceGroup()
+  properties: {
+    roleDefinitionId: subscriptionResourceId('Microsoft.Authorization/roleDefinitions', 'f7669afb-68b2-44b4-9c5f-6d2a47fddda0') // Container Apps SessionPools Contributor
+    principalId: scriptPrincipal!.properties.principalId
+    principalType: 'ServicePrincipal'
+  }
+}
+
+resource deployScript 'Microsoft.Resources/deploymentScripts@2020-10-01' = {
+  name: 'getmcpkey-${suffix}'
+  location: location
+  kind: 'AzureCLI'
+  identity: useManagedIdentity ? {
+    type: 'UserAssigned'
+    userAssignedIdentities: {
+      '${scriptPrincipal!.id}': {}
+    }
+  } : null
+  properties: {
+    azCliVersion: '2.77.0'
+    scriptContent: '''
+      az rest --method post --url "$SESSION_POOL_ID/fetchMCPServerCredentials?api-version=2025-02-02-preview" | jq -c '{"key": .apiKey}' > $AZ_SCRIPTS_OUTPUT_PATH
+    '''
+    timeout: 'PT30M'
+    retentionInterval: 'P1D'
+    cleanupPreference: 'OnSuccess'
+    environmentVariables: [
+      {
+        name: 'SESSION_POOL_ID'
+        value: sessionPool.id
+      }
+    ]
+  }
+}
+
+resource aiAccount 'Microsoft.CognitiveServices/accounts@2025-10-01-preview' = {
+  name: 'aia-${suffix}'
+  location: location
+  kind: 'AIServices'
+  sku: {
+    name: 'S0'
+  }
+  properties: {
+    customSubDomainName: 'myaiaccount-${suffix}'
+    allowProjectManagement: true
+  }
+
+  resource project 'projects' = {
+    name: 'aip-${suffix}s'
+    properties: {
+      description: 'This is my AI project.'
+    }
+
+    resource mcpConn 'connections' = {
+      name: 'aic-${suffix}'
+      properties: {
+        authType: 'CustomKeys'
+        category: 'RemoteTool'
+        credentials: {
+          keys: {
+            'x-ms-apikey': deployScript.properties.outputs.key
+          }
+        }
+        target: sessionPool.properties.mcpServerSettings.mcpServerEndpoint
+      }
+    }
+  }
+
+  resource model 'deployments' = {
+    name: modelDeploymentName
+    sku: {
+      name: 'GlobalStandard'
+      capacity: 1
+    }
+    properties: {
+      model: {
+        format: 'OpenAI'
+        name: modelName
+      }
+    }
+  }
+}
+
+@description('Outputs the ID of the project connection for the Code Interpreter MCP Tool')
+output AZURE_AI_CONNECTION_ID string = aiAccount::project::mcpConn.id
+
+@description('Model deployment name')
+output AZURE_AI_MODEL_DEPLOYMENT_NAME string = aiAccount::model.name
+
+@description('AI Project Endpoint')
+output AZURE_AI_PROJECT_ENDPOINT string = aiAccount::project.properties.endpoints['AI Foundry API']

samples/python/prompt-agents/code-interpreter-custom/main.py

@@ -0,0 +1,60 @@
+import os
+
+import dotenv
+from azure.ai.projects import AIProjectClient
+from azure.ai.projects.models import PromptAgentDefinition, MCPTool
+from azure.identity import DefaultAzureCredential
+
+dotenv.load_dotenv()
+
+project_client = AIProjectClient(
+    endpoint=os.environ["AZURE_AI_PROJECT_ENDPOINT"],
+    credential=DefaultAzureCredential(),
+)
+openai_client = project_client.get_openai_client()
+
+tools = [
+    MCPTool(
+        # This is just a placeholder. Connection details are in
+        # the project connection referenced by `project_connection_id`.
+        server_url="https://localhost",
+        server_label="python_tool",
+        require_approval="never",
+        allowed_tools=[
+            "launchShell",
+            "runPythonCodeInRemoteEnvironment",
+        ],
+        project_connection_id=os.environ["AZURE_AI_CONNECTION_ID"],
+    ),
+]
+
+EXAMPLE_DATA_FILE_URL = "https://raw.githubusercontent.com/Azure-Samples/azureai-samples/refs/heads/main/scenarios/Agents/data/nifty_500_quarterly_results.csv"
+
+with project_client:
+    agent = project_client.agents.create_version(
+        agent_name="MyAgent",
+        definition=PromptAgentDefinition(
+            model=os.environ["AZURE_AI_MODEL_DEPLOYMENT_NAME"],
+            instructions="""\
+You are a helpful agent that can use a Python code interpreter to assist users. Use the `python_tool` MCP
+server to perform any calculations or numerical analyses. ALWAYS call the `launchShell` tool first before
+calling the `runPythonCodeInRemoteEnvironment` tool. If you need to provide any non-text data to the user,
+always print a data URI with the contents. NEVER provide a path to a file in the remote environment to the user.
+""",
+            temperature=0,
+            tools=tools,
+        ),
+    )
+    print(f"Agent created (id: {agent.id}, name: {agent.name}, version: {agent.version})")
+
+    # Use the agent to analyze a CSV file and produce a histogram
+    response = openai_client.responses.create(
+        input=f"Please analyze the CSV file at {EXAMPLE_DATA_FILE_URL}. Could you please create bar chart in the TRANSPORTATION sector for the operating profit and provide a file to me?",
+        extra_body={"agent": {"name": agent.name, "type": "agent_reference"}},
+    )
+    print(f"[Response {response.id}]: {response.output_text}")
+
+    # Clean up resources by deleting the agent version
+    # This prevents accumulation of unused agent versions in your project
+    project_client.agents.delete_version(agent_name=agent.name, agent_version=agent.version)
+    print("Agent deleted")

samples/python/prompt-agents/code-interpreter-custom/pyproject.toml

@@ -0,0 +1,13 @@
+[project]
+name = "code-interpreter-custom"
+version = "0.1.0"
+description = "Basic example for using custom code interpreter session pools."
+readme = "README.md"
+requires-python = ">=3.12"
+dependencies = [
+    "aiohttp>=3.13.2",
+    "azure-ai-projects==2.0.0b2",
+    "azure-identity>=1.25.1",
+    "dotenv>=0.9.9",
+    "openai>=2.8.1",
+]