Configuring Agent Builder

Agent Builder provides comprehensive configuration options for creating production-ready AI agents. This section describes how to configure and manage Agent Builder deployments.

System prompt configuration

The system prompt defines your agent’s behavior, personality, and capabilities. To configure the system prompt:

1. In the Agent Builder wizard, navigate to the Configure Agent step.

2. Edit the system prompt template in the text editor.

3. Include clear instructions for:

   • Agent’s role and purpose

   • How to use available tools (MCP servers)

   • Response formatting preferences

   • Behavioral guidelines

4. Use the Reset to default button to restore the original template if needed.

Best practices for agent prompts:

• Be specific about the agent’s capabilities and limitations

• Provide clear examples of desired behavior

• Include instructions for tool usage and when to invoke them

• Define response format expectations

• Set boundaries for agent behavior

MCP server integration

Model Context Protocol (MCP) servers provide agents with access to enterprise tools and data sources. To configure MCP servers:

1. In the Configure Agent step, locate the MCP Servers section.

2. Select from available MCP servers in the dropdown menu.

Note

MCP servers must be configured and accessible before agent deployment. The agent will automatically discover and use tools exposed by the configured MCP servers. Refer to the MCP documentation for server setup and tool configuration.

Memory settings

Agent Builder provides two types of memory for maintaining context and knowledge:

Short-term memory

Enabled by default for all agents:

• Maintains conversation context within sessions

• Automatically captures user messages and agent responses

• Organized by actorId and sessionId for proper isolation

• No configuration required

Long-term memory

Optional feature for storing insights across sessions:

1. In the Configure Agent step, locate the Memory Configuration section.

2. Toggle Enable long-term memory to activate.

3. When enabled, the agent can:

   • Extract and store important information across conversations

   • Retrieve relevant context from previous sessions

   • Build knowledge about user preferences and history

Note

Long-term memory uses AgentCore Memory with semantic memory strategy and default retention settings.

Monitoring Agent Builder deployments

Agent Builder provides comprehensive monitoring through CloudWatch dashboards and metrics.

Accessing CloudWatch dashboards

1. Navigate to the CloudWatch console in your AWS account.

2. Select Dashboards from the left navigation.

3. Find the dashboard named AgentBuilder-<UseCaseId>.

4. View real-time metrics and historical performance data.

Log access and analysis

Agent logs are available in CloudWatch Logs:

1. Navigate to CloudWatch Logs in the AWS console.

2. Find log groups prefixed with /aws/bedrock-agentcore/runtimes/.

3. Use CloudWatch Insights to query and analyze logs.

4. Search for specific request IDs or error patterns.