Observe by Snowflake Launches MCP Server and CLI for AI Agents

The shift Observe is announcing reflects a fundamental change in how observability platforms must operate: AI agents are now primary consumers of telemetry data alongside humans. Where observability was once a domain exclusively for engineers reading dashboards and logs, agents now investigate errors autonomously, correlate failures across services, and triage alerts before a human is paged. This architectural shift—moving from a UI-centric design to API-first access—is no longer optional.

The removal of the LLM intermediary is the most significant technical change. The old design forced all agent queries through Observe's own reasoning layer, which was costly to operate and gave agents no direct visibility into data structures. Advances in foundation models (Claude, GPT, and others) mean that modern AI agents are sophisticated enough to reason about observability data directly, without needing Observe to interpret the query. By connecting agents to the same APIs that power the Observe UI, the platform gains cost efficiency, reduces latency, and gives agents full autonomy to explore data, query the context graph, and craft precise observability queries. This is also why the CLI and MCP server have full parity: every operation available through the UI is now available programmatically, ensuring no workflow is locked behind a human interface.

The inclusion of prebuilt skills—structured workflows for incident investigation, failure tracing, validation, and outlier detection—lowers the barrier to adoption. Engineers no longer need to write custom agents from scratch; they can compose, automate, and extend common tasks immediately. The real-world use cases mentioned in the article (alert-triage agents that automatically pull production telemetry, copilots that assist during incident investigations, and custom workflows) suggest that the demand for this capability is already present among Snowflake's customer base.