Industry story
Pydantic AI v2.22.0 Adds MCP, Anthropic, and Temporal Fixes
agent-framework mcp orchestration reliability tool-use
Pydantic AI's v2.22.0 release is less about new features and more about making long-running agents production-safe. The Temporal fixes are the substance here: livelock resolution, better serialization error handling, heartbeating in tool and MCP activities. That's the work of a framework that has real production deployments surfacing real failure modes. The MCPToolset additions and Anthropic system-prompt handling are useful, but the hardening of durable execution is what tells you this stack is being pushed past demos.
Full analysis
Pydantic AI released v2.22.0 with several agent-relevant features and bug fixes. New capabilities include: MCPToolset clients can now skip optional MCP (Model Context Protocol — an emerging standard for how agents call external tools) tasks via a prefer_tasks flag; Gemini models get VALIDATED tool mode enabled by default; Anthropic now receives mid-conversation system prompts as native system messages rather than user-turn injections; and a new RunContext.is_tool_available API lets agents introspect which tools are accessible at runtime.
The release is dominated by fixes to Temporal-based durable execution (a workflow engine used to make long-running agent runs reliable and resumable), including livelock resolution, better serialization error handling, heartbeating in tool/MCP activities, and Prefect tool-call cache key corrections. These improvements signal active hardening of production-grade, long-horizon agent orchestration capabilities in the Pydantic AI framework.
Comments