Outil d’interpréteur de code personnalisé pour les agents (préversion)

Exemple de code

L’exemple de Python suivant montre comment créer un agent avec un outil MCP de l’interpréteur de code personnalisé :

from azure.identity import DefaultAzureCredential
from azure.ai.projects import AIProjectClient
from azure.ai.projects.models import MCPTool, MCPToolboxTool, PromptAgentDefinition

# Format: "https://resource_name.ai.azure.com/api/projects/project_name"
PROJECT_ENDPOINT = "your_project_endpoint"
MCP_SERVER_URL = "https://your-mcp-server-url"
# Optional: set to your project connection ID if your MCP server requires authentication
MCP_CONNECTION_ID = "your-mcp-connection-id"

# Create clients to call Foundry API
project = AIProjectClient(
    endpoint=PROJECT_ENDPOINT,
    credential=DefaultAzureCredential(),
)
openai = project.get_openai_client()

# Add the custom code interpreter MCP server to a toolbox. Using a toolbox is the
# recommended way to give agents tools: you curate tools once and reuse the toolbox
# across agents. See /azure/foundry/agents/concepts/toolbox-overview
toolbox = project.toolboxes.create_version(
    name="custom-code-interpreter-toolbox",
    description="Toolbox with the custom code interpreter MCP server",
    tools=[
        MCPToolboxTool(
            server_label="custom-code-interpreter",
            server_url=MCP_SERVER_URL,
            project_connection_id=MCP_CONNECTION_ID,
        )
    ],
)

# The toolbox exposes an MCP-compatible endpoint.
TOOLBOX_MCP_URL = (
    f"{PROJECT_ENDPOINT}/toolboxes/{toolbox.name}"
    f"/versions/{toolbox.version}/mcp?api-version=v1"
)

# Create a remote-tool project connection that points at the toolbox endpoint.
# Use a user Entra token so the caller's identity is passed through
# (audience https://ai.azure.com). Create the connection once, for example with
# the Azure Developer CLI:
#
#    azd ai connection create custom-code-interpreter-toolbox-conn \
#      --kind remote-tool \
#      --target "<TOOLBOX_MCP_URL>" \
#      --auth-type user-entra-token \
#      --audience https://ai.azure.com
TOOLBOX_CONNECTION_NAME = "custom-code-interpreter-toolbox-conn"

# Create an agent that uses the toolbox as an MCP tool
agent = project.agents.create_version(
    agent_name="CustomCodeInterpreterAgent",
    definition=PromptAgentDefinition(
        model="gpt-5-mini",
        instructions="You are a helpful assistant that can run Python code to analyze data and solve problems.",
        tools=[
            MCPTool(
                server_label="toolbox",
                server_url=TOOLBOX_MCP_URL,
                require_approval="never",
                project_connection_id=TOOLBOX_CONNECTION_NAME,
            )
        ],
    ),
    description="Agent with custom code interpreter for data analysis.",
)
print(f"Agent created (id: {agent.id}, name: {agent.name}, version: {agent.version})")

# Test the agent with a simple calculation
response = openai.responses.create(
    input="Calculate the factorial of 10 using Python.",
    extra_body={"agent_reference": {"name": agent.name, "type": "agent_reference"}},
)
print(f"Response: {response.output_text}")

# Clean up
project.agents.delete_version(agent_name=agent.name, agent_version=agent.version)
project.toolboxes.delete_toolbox_version(
  toolbox_name=toolbox.name,
  version=toolbox.version,
)
print("Agent deleted")

Sortie attendue

Lorsque vous exécutez l’exemple, vous voyez une sortie similaire à :

Agent created (id: agent-xxxxxxxxxxxx, name: CustomCodeInterpreterAgent, version: 1)
Response: The factorial of 10 is 3,628,800. I calculated this using Python's math.factorial() function.
Agent deleted

Utiliser un agent hébergé

Cet exemple utilise le FoundryChatClient Microsoft Agent Framework et se connecte au point de terminaison MCP de la boîte à outils à l’aide de FoundryToolbox.

import asyncio

from agent_framework import Agent
from agent_framework.foundry import FoundryChatClient, FoundryToolbox
from azure.identity import AzureCliCredential
from azure.ai.projects import AIProjectClient
from azure.ai.projects.models import MCPToolboxTool

PROJECT_ENDPOINT = "https://<account>.services.ai.azure.com/api/projects/<project>"
MCP_SERVER_URL = "https://your-mcp-server-url"
# Optional: set to your project connection ID if your MCP server requires authentication
MCP_CONNECTION_ID = "your-mcp-connection-id"


async def main() -> None:
    credential = AzureCliCredential()

    # 1. Create the custom code interpreter MCP tool and add it to a toolbox. Using a toolbox is the
    #    recommended way to give agents tools: curate tools once and reuse the
    #    toolbox across agents. See /azure/foundry/agents/concepts/toolbox-overview
    project = AIProjectClient(endpoint=PROJECT_ENDPOINT, credential=credential)
    toolbox = project.toolboxes.create_version(
        name="custom-code-interpreter-toolbox",
        description="Toolbox with the custom code interpreter MCP server",
        tools=[
            MCPToolboxTool(
                server_label="custom-code-interpreter",
                server_url=MCP_SERVER_URL,
                project_connection_id=MCP_CONNECTION_ID,
            )
        ],
    )

    # 2. The toolbox exposes an MCP-compatible endpoint.
    TOOLBOX_MCP_URL = (
        f"{PROJECT_ENDPOINT}/toolboxes/{toolbox.name}"
        f"/versions/{toolbox.version}/mcp?api-version=v1"
    )

    # 3. Attach the toolbox to the hosted agent as an MCP tool.
, timeout=120.0)
    toolbox_tool = FoundryToolbox(credential, url=TOOLBOX_MCP_URL)

agent = Agent(
        client=FoundryChatClient(credential=credential),
        instructions="You are a helpful assistant that can run Python code to analyze data and solve problems.",
        tools=[toolbox_tool],
    )

    result = await agent.run("Calculate the factorial of 10 using Python.")
    print(result.text)


    project.toolboxes.delete_toolbox_version(
      toolbox_name=toolbox.name,
      version=toolbox.version,
    )


if __name__ == "__main__":
    asyncio.run(main())

Exemple de code

L’exemple C# suivant montre comment créer un agent avec un outil MCP de l’interpréteur de code personnalisé. Pour plus d’informations sur l’utilisation des outils MCP dans .NET, consultez l’exemple d’outil MCP dans le Kit de développement logiciel (SDK) Azure pour .NET référentiel sur GitHub.

using System;
using Azure.AI.Projects;
using Azure.AI.Extensions.OpenAI;
using Azure.Identity;

// Format: "https://resource_name.ai.azure.com/api/projects/project_name"
var projectEndpoint = "your_project_endpoint";
var mcpServerUrl = "https://your-mcp-server-url";
// Optional: set to your project connection ID if your MCP server requires authentication
var mcpConnectionId = "your-mcp-connection-id";

// Create project client to call Foundry API
AIProjectClient projectClient = new(
    endpoint: new Uri(projectEndpoint),
    tokenProvider: new DefaultAzureCredential());

// Add the custom code interpreter MCP server to a toolbox. Using a toolbox is the
// recommended way to give agents tools. See /azure/foundry/agents/concepts/toolbox-overview
// Code runs in a sandboxed Azure Container Apps session.
McpTool customCodeInterpreter = ResponseTool.CreateMcpTool(
    serverLabel: "custom-code-interpreter",
    serverUri: new Uri(mcpServerUrl));
customCodeInterpreter.ProjectConnectionId = mcpConnectionId;

ToolboxVersion toolboxVersion = projectClient.AgentAdministrationClient
    .GetAgentToolboxes().CreateToolboxVersion(
        toolboxName: "custom-code-interpreter-toolbox",
        tools: [ProjectsAgentTool.AsProjectTool(customCodeInterpreter)],
        description: "Toolbox with the custom code interpreter MCP server");

// The toolbox exposes an MCP-compatible endpoint.
var toolboxMcpUrl = new Uri(
    $"{projectEndpoint}/toolboxes/{toolboxVersion.Name}" +
    $"/versions/{toolboxVersion.Version}/mcp?api-version=v1");

// Create a remote-tool project connection that points at the toolbox endpoint.
// Use a user Entra token so the caller's identity is passed through
// (audience https://ai.azure.com). Create the connection once, for example
// with the Azure Developer CLI:
//
//    azd ai connection create custom-code-interpreter-toolbox-conn \
//      --kind remote-tool \
//      --target "<toolboxMcpUrl>" \
//      --auth-type user-entra-token \
//      --audience https://ai.azure.com
var toolboxConnectionName = "custom-code-interpreter-toolbox-conn";

McpTool toolboxTool = ResponseTool.CreateMcpTool(
    serverLabel: "toolbox",
    serverUri: toolboxMcpUrl,
    toolCallApprovalPolicy: new McpToolCallApprovalPolicy(
        GlobalMcpToolCallApprovalPolicy.NeverRequireApproval));
toolboxTool.ProjectConnectionId = toolboxConnectionName;

DeclarativeAgentDefinition agentDefinition = new(model: "gpt-5-mini")
{
    Instructions = "You are a helpful assistant that can run Python code to analyze data and solve problems.",
    Tools = { toolboxTool }
};

AgentVersion agent = projectClient.AgentAdministrationClient.CreateAgentVersion(
    agentName: "CustomCodeInterpreterAgent",
    options: new(agentDefinition));

Console.WriteLine($"Agent created: {agent.Name} (version {agent.Version})");

// Create a response using the agent
ProjectResponsesClient responseClient = projectClient.ProjectOpenAIClient.GetProjectResponsesClientForAgent(agent.Name);

ResponseResult response = responseClient.CreateResponse(
    new([ResponseItem.CreateUserMessageItem("Calculate the factorial of 10 using Python.")]));

Console.WriteLine(response.GetOutputText());

// Clean up
projectClient.AgentAdministrationClient.DeleteAgentVersion(
    agentName: agent.Name,
    agentVersion: agent.Version);
Console.WriteLine("Agent deleted");

Supprimez la version de la boîte à outils après que l’agent ne le référence plus. Consultez Supprimer une version de boîte à outils pour l’appel .NET vérifié.

Sortie attendue

Agent created: CustomCodeInterpreterAgent (version 1)
The factorial of 10 is 3,628,800.
Agent deleted

Utiliser un agent hébergé

Cet exemple utilise l’intégration Microsoft Agent Framework AddFoundryToolboxes pour connecter l’agent hébergé à la boîte à outils.

using System;
using Azure.AI.AgentServer.Responses;
using Azure.AI.AgentServer.Responses.Models;
using Azure.AI.OpenAI;
using Azure.AI.Projects;
using Azure.AI.Extensions.OpenAI;
using Azure.Identity;
using Microsoft.Agents.AI;
using Microsoft.Agents.AI.Foundry.Hosting;
using Microsoft.Extensions.DependencyInjection;
using OpenAI.Chat;

const string AgentInstructions = "You are a helpful assistant that can run Python code to analyze data and solve problems.";
const string AgentName = "CustomCodeInterpreterAgent";

string projectEndpoint = Environment.GetEnvironmentVariable("AZURE_AI_PROJECT_ENDPOINT")
    ?? "https://<account>.services.ai.azure.com/api/projects/<project>";
string openAiEndpoint = Environment.GetEnvironmentVariable("AZURE_OPENAI_ENDPOINT")
    ?? throw new InvalidOperationException("AZURE_OPENAI_ENDPOINT is not set.");
string deploymentName = Environment.GetEnvironmentVariable("AZURE_AI_MODEL_DEPLOYMENT_NAME") ?? "gpt-5-mini";
string mcpServerUrl = "https://your-mcp-server-url";
string mcpConnectionId = "your-mcp-connection-id";

DefaultAzureCredential credential = new();

// 1. Create the custom code interpreter MCP tool and add it to a toolbox. Using a toolbox is the
//    recommended way to give agents tools. See /azure/foundry/agents/concepts/toolbox-overview
AIProjectClient projectClient = new(
    endpoint: new Uri(projectEndpoint),
    tokenProvider: credential);
McpTool customCodeInterpreter = ResponseTool.CreateMcpTool(
    serverLabel: "custom-code-interpreter",
    serverUri: new Uri(mcpServerUrl));
customCodeInterpreter.ProjectConnectionId = mcpConnectionId;
ToolboxVersion toolboxVersion = projectClient.AgentAdministrationClient
    .GetAgentToolboxes().CreateToolboxVersion(
        toolboxName: "custom-code-interpreter-toolbox",
        tools: [ProjectsAgentTool.AsProjectTool(customCodeInterpreter)],
        description: "Toolbox with the custom code interpreter MCP server");

// Create the hosted agent and register the toolbox integration.
AIAgent agent = projectClient.AsAIAgent(
    model: deploymentName,
    instructions: "You are a helpful assistant with access to the toolbox tools.",
    name: "hosted-toolbox-agent");

var builder = WebApplication.CreateBuilder(args);
builder.Services.AddFoundryResponses(agent);
builder.Services.AddFoundryToolboxes(credential, toolboxVersion.Name);

var app = builder.Build();
app.MapFoundryResponses();
app.Run();

Exemple de code

L’exemple TypeScript suivant montre comment créer un agent avec un outil MCP de l’interpréteur de code personnalisé. Pour obtenir une version javaScript, consultez l’exemple d’outil MCP dans le Kit de développement logiciel (SDK) Azure du référentiel JavaScript sur GitHub.

import { DefaultAzureCredential } from "@azure/identity";
import { AIProjectClient } from "@azure/ai-projects";

// Format: "https://resource_name.ai.azure.com/api/projects/project_name"
const PROJECT_ENDPOINT = "your_project_endpoint";
const MCP_SERVER_URL = "https://your-mcp-server-url";

export async function main(): Promise<void> {
  // Create clients to call Foundry API
  const project = new AIProjectClient(PROJECT_ENDPOINT, new DefaultAzureCredential());
  const openai = project.getOpenAIClient();

  // Add the custom code interpreter MCP server to a toolbox. Using a toolbox is
  // the recommended way to give agents tools. Code runs in a sandboxed Azure
  // Container Apps session, so the tool uses require_approval: "never".
  // See /azure/foundry/agents/concepts/toolbox-overview
  const toolbox = await project.toolboxes.createVersion(
    "custom-code-interpreter-toolbox",
    [
      {
        type: "mcp",
        server_label: "custom-code-interpreter",
        server_url: MCP_SERVER_URL,
        require_approval: "never",
      },
    ],
    { description: "Toolbox with the custom code interpreter MCP server" },
  );

  // The toolbox exposes an MCP-compatible endpoint.
  const toolboxMcpUrl =
    `${PROJECT_ENDPOINT}/toolboxes/${toolbox.name}` +
    `/versions/${toolbox.version}/mcp?api-version=v1`;

  // Create a remote-tool project connection that points at the toolbox endpoint.
  // Use a user Entra token so the caller's identity is passed through
  // (audience https://ai.azure.com). Create the connection once, for example
  // with the Azure Developer CLI:
  //
  //    azd ai connection create custom-code-interpreter-toolbox-conn \
  //      --kind remote-tool \
  //      --target "<toolboxMcpUrl>" \
  //      --auth-type user-entra-token \
  //      --audience https://ai.azure.com
  const toolboxConnectionName = "custom-code-interpreter-toolbox-conn";

  // Create an agent that uses the toolbox as an MCP tool
  const agent = await project.agents.createVersion("CustomCodeInterpreterAgent", {
    kind: "prompt",
    model: "gpt-5-mini",
    instructions:
      "You are a helpful assistant that can run Python code to analyze data and solve problems.",
    tools: [
      {
        type: "mcp",
        server_label: "toolbox",
        server_url: toolboxMcpUrl,
        require_approval: "never",
        project_connection_id: toolboxConnectionName,
      },
    ],
  });
  console.log(`Agent created (name: ${agent.name}, version: ${agent.version})`);

  // Send a request to the agent
  const response = await openai.responses.create(
    {
      input: "Calculate the factorial of 10 using Python.",
    },
    {
      body: { agent_reference: { name: agent.name, type: "agent_reference" } },
    },
  );
  console.log(`Response: ${response.output_text}`);

  // Clean up
  await project.agents.deleteVersion(agent.name, agent.version);
  await project.toolboxes.deleteVersion(toolbox.name, toolbox.version);
  console.log("Agent deleted");
}

main().catch((err) => {
  console.error("The sample encountered an error:", err);
});

Sortie attendue

Agent created (name: CustomCodeInterpreterAgent, version: 1)
Response: The factorial of 10 is 3,628,800. I calculated this using Python's math.factorial() function.
Agent deleted

Ajoutez la dépendance à votre pom.xml:

<dependency>
    <groupId>com.azure</groupId>
    <artifactId>azure-ai-agents</artifactId>
    <version>2.4.0</version>
</dependency>

Exemple de code

import com.azure.ai.agents.AgentsClient;
import com.azure.ai.agents.AgentsClientBuilder;
import com.azure.ai.agents.ResponsesClient;
import com.azure.ai.agents.models.AgentReference;
import com.azure.ai.agents.models.AgentVersionDetails;
import com.azure.ai.agents.models.AzureCreateResponseOptions;
import com.azure.ai.agents.models.McpTool;
import com.azure.ai.agents.models.PromptAgentDefinition;
import com.azure.identity.DefaultAzureCredentialBuilder;
import com.openai.models.responses.Response;
import com.openai.models.responses.ResponseCreateParams;

import java.util.Collections;

public class CustomCodeInterpreterExample {
    public static void main(String[] args) {
        // Format: "https://resource_name.ai.azure.com/api/projects/project_name"
        String projectEndpoint = "your_project_endpoint";
        String toolboxMcpUrl = projectEndpoint + "/toolboxes/custom-code-interpreter-toolbox/versions/1/mcp?api-version=v1";
        // Set to the remote-tool project connection that points at the toolbox MCP endpoint.
        String toolboxConnectionId = "custom-code-interpreter-toolbox-conn";

        // Create clients to call Foundry API
        AgentsClientBuilder builder = new AgentsClientBuilder()
            .credential(new DefaultAzureCredentialBuilder().build())
            .endpoint(projectEndpoint);

        AgentsClient agentsClient = builder.buildAgentsClient();
        ResponsesClient responsesClient = builder.buildResponsesClient();

        // Attach the toolbox MCP endpoint as an MCP tool.
        // Uses require_approval: "never" because code runs in a sandboxed Container Apps session.
        McpTool toolboxTool = new McpTool("toolbox")
            .setServerUrl(toolboxMcpUrl)
            .setProjectConnectionId(toolboxConnectionId)
            .setRequireApproval("never");

        PromptAgentDefinition agentDefinition = new PromptAgentDefinition("gpt-5-mini")
            .setInstructions("You are a helpful assistant that can run Python code to analyze data and solve problems.")
            .setTools(Collections.singletonList(toolboxTool));

        AgentVersionDetails agent = agentsClient.createAgentVersion(
            "CustomCodeInterpreterAgent", agentDefinition);
        System.out.printf("Agent created: %s (version %s)%n", agent.getName(), agent.getVersion());

        // Create a response
        AgentReference agentReference = new AgentReference(agent.getName())
            .setVersion(agent.getVersion());

        Response response = responsesClient.createAzureResponse(
            new AzureCreateResponseOptions().setAgentReference(agentReference),
            ResponseCreateParams.builder()
                .input("Calculate the factorial of 10 using Python."));

        System.out.println("Response: " + response.output());

        // Clean up
        agentsClient.deleteAgentVersion(agent.getName(), agent.getVersion());
        System.out.println("Agent deleted");
    }
}

Sortie attendue

Agent created: CustomCodeInterpreterAgent (version 1)
Response: The factorial of 10 is 3,628,800.
Agent deleted

Conditions préalables

Définissez ces variables d’environnement :

• FOUNDRY_PROJECT_ENDPOINT: URL du point de terminaison de votre projet.
• AGENT_TOKEN : jeton du porteur pour Foundry.

Obtenez un jeton d’accès :

export AGENT_TOKEN=$(az account get-access-token --scope "https://ai.azure.com/.default" --query accessToken -o tsv)

Exemple de code

Créer une boîte à outils avec l’interpréteur de code personnalisé

Ajoutez l’interpréteur de code personnalisé en créant une boîte à outils. Ensuite, attachez la boîte à outils à votre agent en tant qu’outil MCP. Pour plus d’informations, consultez Qu’est-ce qu’une boîte à outils ?

curl -X POST "$FOUNDRY_PROJECT_ENDPOINT/toolboxes/custom-code-interpreter-toolbox/versions?api-version=v1" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AGENT_TOKEN" \
  -d '{
    "description": "Toolbox with the custom code interpreter MCP server",
    "tools": [
      {
        "type": "mcp",
        "server_label": "custom-code-interpreter",
        "server_url": "<MCP_SERVER_URL>",
        "project_connection_id": "<MCP_PROJECT_CONNECTION_ID>",
        "require_approval": "never"
      }
    ]
  }'

La boîte à outils expose un point de terminaison compatible MCP à l’emplacement $FOUNDRY_PROJECT_ENDPOINT/toolboxes/custom-code-interpreter-toolbox/versions/<version>/mcp?api-version=v1où <version> est la version retournée par l’appel précédent.

Créer une connexion à l’outil distant pour la boîte à outils

Créez une connexion de projet d’outil distant qui pointe vers le point de terminaison de la boîte à outils. Utilisez un jeton Entra utilisateur pour que l’identité de l’appelant soit transmise (audience https://ai.azure.com) :

azd ai connection create custom-code-interpreter-toolbox-conn \
  --kind remote-tool \
  --target "$FOUNDRY_PROJECT_ENDPOINT/toolboxes/custom-code-interpreter-toolbox/versions/<version>/mcp?api-version=v1" \
  --auth-type user-entra-token \
  --audience https://ai.azure.com

Créer un agent qui utilise la boîte à outils

curl -X POST "$FOUNDRY_PROJECT_ENDPOINT/agents?api-version=v1" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AGENT_TOKEN" \
  -d '{
    "name": "CustomCodeInterpreterAgent",
    "definition": {
      "kind": "prompt",
      "model": "<MODEL_DEPLOYMENT>",
      "instructions": "You are a helpful assistant that can run Python code to analyze data and solve problems.",
      "tools": [
        {
          "type": "mcp",
          "server_label": "toolbox",
          "server_url": "'$FOUNDRY_PROJECT_ENDPOINT'/toolboxes/custom-code-interpreter-toolbox/versions/<version>/mcp?api-version=v1",
          "require_approval": "never",
          "project_connection_id": "custom-code-interpreter-toolbox-conn"
        }
      ]
    }
  }'

Créer une réponse

curl -X POST "$FOUNDRY_PROJECT_ENDPOINT/openai/v1/responses" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AGENT_TOKEN" \
  -d '{
    "agent_reference": {"type": "agent_reference", "name": "CustomCodeInterpreterAgent"},
    "input": "Calculate the factorial of 10 using Python."
  }'

Nettoyer

curl -X DELETE "$FOUNDRY_PROJECT_ENDPOINT/agents/CustomCodeInterpreterAgent?api-version=v1" \
  -H "Authorization: Bearer $AGENT_TOKEN"

curl -X DELETE \
  "$FOUNDRY_PROJECT_ENDPOINT/toolboxes/custom-code-interpreter-toolbox/versions/<version>?api-version=v1" \
  -H "Authorization: Bearer $AGENT_TOKEN"

Sortie attendue

{
  "id": "resp_xxxxxxxxxxxx",
  "output": [
    {
      "type": "message",
      "role": "assistant",
      "content": [
        {
          "type": "output_text",
          "text": "The factorial of 10 is 3,628,800."
        }
      ]
    }
  ]
}