Introducing stateful MCP client capabilities on Amazon Bedrock AgentCore Runtime

This article introduces stateful Model Context Protocol (MCP) client capabilities on Amazon Bedrock AgentCore Runtime, enabling interactive multi-turn agent workflows.

• Stateful mode enables three new client capabilities: elicitation, sampling, and progress notifications
• Elicitation allows servers to pause execution and request structured user input mid-workflow
• Sampling enables servers to request LLM-generated content from the client without holding model credentials
• Progress notifications stream real-time updates during long-running operations via fire-and-forget events
• Stateful mode provisions dedicated microVMs per session with up to 8 hours lifetime or 15-minute idle timeout
• Configuration requires single flag change: stateless_http=False in server startup
• Includes code examples for expense tracking, spending analysis, and financial report generation
• All capabilities are opt-in; clients declare support during initialization handshake

Stateful MCP transforms one-way tool execution into bidirectional conversations, enabling agents to handle interactive workflows, leverage client LLMs, and provide real-time operation visibility on managed infrastructure.