migration changedmanaged-agents/migration

The first capture of this source. The page was already there, and this is what it said.

---
title: Migration
url: https://platform.claude.com/docs/en/managed-agents/migration
description: Move an existing agent built on the Messages API or the Claude Agent SDK to Claude Managed Agents.
---

Claude Managed Agents replaces your hand-written agent loop with managed infrastructure. This page covers what changes when you migrate from a custom loop built on the [Messages API](https://platform.claude.com/docs/en/build-with-claude/working-with-messages) or from the [Claude Agent SDK](https://code.claude.com/docs/en/agent-sdk/overview).

<Note>
  Managed Agents API requests require the `managed-agents-2026-04-01` beta header, except memory store endpoints, which use `agent-memory-2026-07-22` instead. The SDK sets the correct beta header automatically. See [Beta headers](https://platform.claude.com/docs/en/api/beta-headers#endpoint-specific-headers).
</Note>

## From a Messages API agent loop

If you built an agent by calling `messages.create` in a `while` loop, running tool calls yourself, and appending results to the conversation history, most of that code goes away.

### What you stop managing

| Before                                                                                           | After                                                                                                                      |
| ------------------------------------------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------------- |
| You maintain the conversation history array and pass it back on every turn.                      | The session stores history server-side. Send events, receive events.                                                       |
| You iterate `tool_use` content blocks, run each tool, and loop back with `tool_result` messages. | Pre-built tools run inside the sandbox automatically. You only handle custom tools through `agent.custom_tool_use` events. |
| You provision your own sandbox for running agent-generated code.                                 | The session sandbox handles code execution, file operations, and bash.                                                     |
| You decide when the loop is done.                                                                | The session emits `session.status_idle` when the agent has nothing more to do.                                             |

### Code comparison

**Before** (Messages API loop, simplified):

<CodeGroup>
  ```python Python
  messages = [{"role": "user", "content": task}]
  while True:
      response = client.messages.create(
          model="claude-opus-5",
          max_tokens=1024,
          messages=messages,
          tools=tools,
      )
      messages.append({"role": "assistant", "content": response.content})
      if response.stop_reason == "end_turn":
          break
      for block in response.content:
          if block.type == "tool_use":
              result = execute_tool(block.name, block.input)
              messages.append(
                  {
                      "role": "user",
                      "content": [
                          {
                              "type": "tool_result",
                              "tool_use_id": block.id,
                              "content": result,
                          }
                      ],
                  }
              )
  ```

  ```typescript TypeScript
  const messages: Anthropic.MessageParam[] = [{ role: "user", content: task }];
  while (true) {
    const response = await client.messages.create({
      model: "claude-opus-5",
      max_tokens: 1024,
      messages,
      tools
    });
    messages.push({ role: "assistant", content: response.content });
    if (response.stop_reason === "end_turn") {
      break;
    }
    for (const block of response.content) {
      if (block.type === "tool_use") {
        const result = executeTool(block.name, block.input);
        messages.push({
          role: "user",
          content: [
            {
              type: "tool_result",
              tool_use_id: block.id,
              content: result
            }
          ]
        });
      }
    }
  }
  ```

  ```csharp C#
  List<MessageParam> messages = [new() { Role = Role.User, Content = task }];
  while (true)
  {
      var response = await client.Messages.Create(new()
      {
          Model = Model.ClaudeOpus5,
          MaxTokens = 1024,
          Messages = messages,
          Tools = tools,
      });
      messages.Add(new()
      {
          Role = Role.Assistant,
          Content = new([.. response.Content.Select(block => new ContentBlockParam(block.Json))]),
      });
      if (response.StopReason == StopReason.EndTurn)
      {
          break;
      }
      foreach (var block in response.Content)
      {
          if (block.Value is ToolUseBlock toolUse)
          {
              var result = ExecuteTool(toolUse.Name, toolUse.Input);
              messages.Add(new()
              {
                  Role = Role.User,
                  Content = new([new ToolResultBlockParam { ToolUseID = toolUse.ID, Content = result }]),
              });
          }
      }
  }
  ```

  ```go Go
  messages := []anthropic.MessageParam{
  	anthropic.NewUserMessage(anthropic.NewTextBlock(task)),
  }
  for {
  	response, err := client.Messages.New(ctx, anthropic.MessageNewParams{
  		Model:     anthropic.ModelClaudeOpus5,
  		MaxTokens: 1024,
  		Messages:  messages,
  		Tools:     tools,
  	})
  	if err != nil {
  		log.Fatal(err)
  	}
  	messages = append(messages, response.ToParam())
  	if response.StopReason == anthropic.StopReasonEndTurn {
  		break
  	}
  	for _, block := range response.Content {
  		if toolUse, ok := block.AsAny().(anthropic.ToolUseBlock); ok {
  			result := executeTool(toolUse.Name, toolUse.Input)
  			messages = append(messages, anthropic.NewUserMessage(
  				anthropic.NewToolResultBlock(toolUse.ID, result, false),
  			))
  		}
  	}
  }
  ```

  ```java Java
  var messages = new ArrayList<MessageParam>();
  messages.add(MessageParam.builder()
      .role(MessageParam.Role.USER)
      .content(task)
      .build());
  while (true) {
      var response = client.messages().create(MessageCreateParams.builder()
          .model(Model.CLAUDE_OPUS_5)
          .maxTokens(1024)
          .messages(messages)
          .tools(tools)
          .build());
      messages.add(response.toParam());
      if (StopReason.END_TURN.equals(response.stopReason().orElse(null))) {
          break;
      }
      for (var block : response.content()) {
          block.toolUse().ifPresent(toolUse -> {
              var result = executeTool(toolUse.name(), toolUse._input());
              messages.add(MessageParam.builder()
                  .role(MessageParam.Role.USER)
                  .contentOfBlockParams(List.of(
                      ContentBlockParam.ofToolResult(ToolResultBlockParam.builder()
                          .toolUseId(toolUse.id())
                          .content(result)
                          .build())))
                  .build());
          });
      }
  }
  ```

  ```php PHP
  $messages = [['role' => 'user', 'content' => $task]];
  while (true) {
      $response = $client->messages->create(
          model: 'claude-opus-5',
          maxTokens: 1024,
          messages: $messages,
          tools: $tools,
      );
      $messages[] = ['role' => 'assistant', 'content' => $response->content];
      if ($response->stopReason === 'end_turn') {
          break;
      }
      foreach ($response->content as $block) {
          if ($block->type === 'tool_use') {
              $result = executeTool($block->name, $block->input);
              $messages[] = [
                  'role' => 'user',
                  'content' => [
                      [
                          'type' => 'tool_result',
                          'tool_use_id' => $block->id,
                          'content' => $result,
                      ],
                  ],
              ];
          }
      }
  }
  ```

  ```ruby Ruby
  messages = [{ role: "user", content: task }]
  loop do
    response = client.messages.create(
      model: "claude-opus-5",
      max_tokens: 1024,
      messages: messages,
      tools: tools
    )
    messages << { role: "assistant", content: response.content }
    break if response.stop_reason == :end_turn
    response.content.each do |block|
      next unless block.type == :tool_use
      result = execute_tool(block.name, block.input)
      messages << {
        role: "user",
        content: [
          {
            type: "tool_result",
            tool_use_id: block.id,
            content: result
          }
        ]
      }
    end
  end
  ```
</CodeGroup>

**After** (Claude Managed Agents):

<CodeGroup>
  ```bash cURL
  agent=$(
    curl --fail-with-body -sS "https://api.anthropic.com/v1/agents?beta=true" \
      -H "x-api-key: ${ANTHROPIC_API_KEY}" \
      -H "anthropic-version: 2023-06-01" \
      -H "anthropic-beta: managed-agents-2026-04-01" \
      --json '{
        "name": "Task Runner",
        "model": "claude-opus-5",
        "tools": [{"type": "agent_toolset_20260401"}]
      }'
  )
  agent_id=$(jq -r '.id' <<< "${agent}")

  session_id=$(
    curl --fail-with-body -sS "https://api.anthropic.com/v1/sessions?beta=true" \
      -H "x-api-key: ${ANTHROPIC_API_KEY}" \
      -H "anthropic-version: 2023-06-01" \
      -H "anthropic-beta: managed-agents-2026-04-01" \
      --json "$(jq -n --argjson a "${agent}" --arg env "${environment_id}" \
        '{agent: {type: "agent", id: $a.id, version: $a.version}, environment_id: $env}')" \
    | jq -r '.id'
  )

  # Open the SSE stream in the background, then send the user message.
  stream_log=$(mktemp)
  curl --fail-with-body -sS -N \
    "https://api.anthropic.com/v1/sessions/${session_id}/events/stream?beta=true" \
    -H "x-api-key: ${ANTHROPIC_API_KEY}" \
    -H "anthropic-version: 2023-06-01" \
    -H "anthropic-beta: managed-agents-2026-04-01" \
    > "${stream_log}" &
  stream_pid=$!

  curl --fail-with-body -sS \
    "https://api.anthropic.com/v1/sessions/${session_id}/events?beta=true" \
    -H "x-api-key: ${ANTHROPIC_API_KEY}" \
    -H "anthropic-version: 2023-06-01" \
    -H "anthropic-beta: managed-agents-2026-04-01" \
    --json "$(jq -n --arg text "${task}" \
      '{events: [{type: "user.message", content: [{type: "text", text: $text}]}]}')" \
    > /dev/null

  # Wait for the session to go idle. grep exits at the first match, and
  # reading via process substitution means the shell doesn't wait for
  # tail (a foreground `tail -f | grep -m1` pipeline would hang: tail
  # only dies on its next write, which never comes once the stream is idle).
  grep -m1 '"session.status_idle"' <(tail -f -n +1 "${stream_log}") > /dev/null

  kill "${stream_pid}" 2>/dev/null || true

Cut at 300 lines. The page has the rest.

No line in this hunk matches that.