Observe by Snowflake launches MCP server and CLI for AI agents to query telemetry

The shift toward AI agents consuming telemetry represents a fundamental change in how observability platforms function. Historically, telemetry systems were built primarily for human operators reading dashboards and logs; the new requirement is that these systems also expose data to AI agents through programmatic interfaces. Snowflake's announcement reflects this reality: engineers are already building custom alert-triage agents and incident-investigation copilots, but the Observe platform was not originally designed to serve them efficiently.

The economic architecture of the old MCP server illustrates why this redesign matters. By embedding an LLM intermediary that interpreted agent queries before returning answers, Snowflake was duplicating reasoning work—the agent would form a question, send it to Observe's LLM, wait for an interpretation and answer, then potentially ask a follow-up. The new design drops this middleman, letting agents connect directly to Observe's backend APIs and reason over raw data themselves. This reduces both latency (fewer network hops) and cost (no redundant LLM call per query). The body notes that several engineering teams had already built workarounds to bypass this inefficiency, underscoring the market demand for direct access.

The parity between the MCP server and the new CLI is a pragmatic choice. By ensuring every operation available to agents through MCP is also available as a command-line command, Snowflake enables the same custom workflows to run in interactive terminal sessions, within agent environments like Claude Code, or even autonomously in the background without an engineer present. This flexibility supports both immediate incident response (an engineer guiding an investigation) and routine background tasks (autonomous error triage), broadening the platform's utility.