Alibaba contributed OpenSandbox, a secure sandbox for running agents, and OpenAI contributed Open Responses, a specification for interoperable LLM interfaces. Anthropic added Agent Skills, and Hugging Face added OpenEnv, a framework for reinforcement learning environments.
Anthropic's Den Delimarsky, the protocol's lead maintainer, said cumulative MCP SDK downloads have reached 4.17 billion. About a third of the 116 conformance scenarios now test security, and enterprise-managed authorization is live in Claude. His auth roadmap puts proof-of-possession tokens and workload identity federation in draft.
OpenAI's Thomas Sarlandie said the ChatGPT plugin directory rejects submitted servers unless every tool carries annotations. Chrome's Sarah Drasner presented WebMCP, a proposed web standard in origin trial that lets websites expose tools to agents, citing benchmarks of 2.5 to 7.5 times faster actions. Several figures presented on stage were not backed by published data.
Why it matters if you build agents
If you run MCP servers, annotate every tool and keep tool schemas small, since oversized schemas get trimmed starting with argument descriptions. Plan for agents that carry their own identity instead of a borrowed user token. Some of the security extensions discussed are still drafts, so treat them as unfinished.
Get stories like this every weekday
One short email with the day's agent tooling news.