The Claude Agent SDK now runs MCP tasks in the background when CLAUDE_AUTO_BACKGROUND_TASKS=1 with streaming input, so long-running tool calls no longer block the session. Not available on Bedrock, Vertex or Foundry.
Key takeaways · AI-distilled
The TypeScript AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → SDK v0.3.319 adds MCPThe Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.Full definition → task support in SDK sessions: with CLAUDE_AUTO_BACKGROUND_TASKS=1 and streamingSending a model's response token by token as it is generated, so the reader sees text immediately instead of waiting for the whole answer.Full definition → input, an MCP task runs in the background as task_type "mcp_task". The release says this is not available on Bedrock, Vertex or Foundry.
The OpenTelemetry claude_code.session.count metric no longer counts prewarm() processes that were never claimed. A pre-started process now counts once, when SpareProcess.claim() succeeds.
session_writes_lost is now sent only when the CLI exits 0, so a run ending in an error result gets only the stderr line, and deleteSession() now rejects with ENOTEMPTY when it cannot fully delete the session folder.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
MCP — The Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.
streaming — Sending a model's response token by token as it is generated, so the reader sees text immediately instead of waiting for the whole answer.
