1.10.0

CrewAI Inc. has launched version 1.10.0 of their popular open-source framework for orchestrating role-playing AI agents, marking a substantial update focused on human oversight and tool reliability. The release, led by maintainer greysonlalonde with contributions from 18 developers, introduces enhanced Human-in-the-Loop (HITL) functionality that enables self-looping feedback mechanisms, allowing agents to pause execution and request human input during complex workflows. This addresses a critical need for oversight in autonomous systems. Additionally, the update significantly improves Model Context Protocol (MCP) tool resolution and event handling, ensuring AI agents can more reliably discover and utilize external tools and data sources.

The technical overhaul includes migrating the CLI's HTTP client from the requests library to httpx for better async performance, adding versioned documentation, and implementing yanked version detection to prevent broken dependencies. Under the hood, the team refactored the core 'crew' and HITL systems into provider patterns for cleaner architecture and improved type safety. The release also patches numerous bugs, including fixing cyclic flows that could silently break, resolving race conditions in guardrail tests, and ensuring OpenAI tool call streams are properly finalized. With 96 new actions added across 9 integrations and enhanced JSON argument validation, version 1.10.0 represents CrewAI's continued evolution from a simple orchestration tool toward a robust platform for building controllable, production-grade AI agent systems.