Snowflake's Observe adds AI agent access via MCP server and CLI

Observability platforms have traditionally been built for human users navigating UIs and dashboards. Snowflake's announcement reflects a fundamental shift: as AI agents take on operational responsibilities—triaging alerts, investigating incidents, correlating failures—observability tools must be redesigned to serve both humans and machines. The old Observe MCP server architecture included an internal LLM to interpret agent queries, a design that made sense when agents and foundation models were less capable. With advances in foundation models, that middleman became unnecessary overhead: agents now understand context graphs and can write precise queries themselves.

The removal of the LLM intermediary is not merely a technical optimization; it solves a practical cost and latency problem that engineering teams had already worked around. By giving agents direct access to the same APIs powering the Observe UI, the new architecture enables faster queries, lower operational costs per call, and more accurate results—agents can examine the data structure and choose which datasets to query before formulating a response. The addition of prebuilt skills (structured workflows for incident investigation, failure tracing, and outlier detection) lowers the barrier to entry: teams can start building agent-driven observability workflows immediately without custom setup.