strands-agents/sdk-typescript ? reverse-engineered prompt

Build me a TypeScript SDK for making AI agents that feels simple to use but still powerful.

I want to be able to create an agent, send it a prompt, and get a response back, with support for streaming as the answer comes in. It should work with both Amazon Bedrock and OpenAI, and make it easy to swap models without rewriting the app. Please include a clean way to add tools with typed inputs, plus structured output so the agent can return validated JSON shaped by a schema. If the model gives bad output, retry automatically and surface a clear error if it still fails.

Also add support for connecting to MCP servers, some basic conversation history handling, and a way to hook into the agent lifecycle for logging or custom behavior. If it makes sense, include a simple example app or demo so I can see how to use it right away. Look up the current docs online if you need to.