Snowflake launches AI agent tools for observability data

Snowflake has announced the general availability of two new tools for connecting AI agents to its Observe observability platform: a redesigned MCP (model context protocol) server and a new CLI (command line interface) that provide agents with programmatic access to production telemetry data. General availability is rolling out now across all clusters, with eu-2 and ca-1 regions to follow shortly.

The key innovation is architectural. The previous Observe MCP server included an internal LLM that received queries from agents, performed the reasoning, and returned answers. This design created two problems: it was expensive to operate (Observe bore the cost of running an LLM for every agent query) and it restricted what agents could do (they had no direct visibility into the data structure and could not explore datasets independently). The new MCP server removes this middleman entirely. Agents now connect directly to Observe's APIs—the same APIs that power the web frontend—giving them direct access to the context graph, APM service maps, OpenTelemetry data, and active alerts. This eliminates both the latency and cost of the old design while enabling agents to reason about the data more effectively, since they can determine which datasets are most valuable and write efficient queries directly.

The new CLI provides the same surface. It gives engineers and agents programmatic access to Observe's full platform from the command line, working from interactive terminal sessions as well as agent environments like Claude Code. The previous CLI covered only a narrow slice of the platform's capabilities; the new one provides access to everything available in the UI and MCP server, allowing both humans and machines to compose, automate, and extend observability workflows.

Both tools ship with prebuilt skills—structured workflows built on how Observe engineers solve common observability tasks. These include investigating incidents, tracing failures, validating changes, finding n+1 issues, and detecting outliers. Agents and engineers can use these skills directly without additional setup after initial configuration. In practice, teams are already using the tools to build custom alert-triage agents that automatically query Observe when an alert fires and pull production context for debugging, as well as copilots that work alongside humans during incident investigations. Some workflows run autonomously in the background to handle routine tasks; others are designed for interactive use, requiring an engineer to guide the investigation.

Access is straightforward: engineers log into their Observe tenant, navigate to "Manage account" → "MCP server details", and follow setup instructions for their agent of choice. Snowflake frames this release as part of a shift in how observability is consumed—from a data problem solved primarily by humans reading dashboards to one where AI agents read and act on telemetry directly, making observability accessible to every tool in a development stack.