Observe connects AI agents to telemetry

Snowflake is making its observability platform, Observe, more accessible to artificial intelligence agents. The company announced the general availability of a redesigned Model Context Protocol (MCP) server and a new command-line interface (CLI) tool. These updates aim to allow AI agents direct access to operational telemetry data, bypassing traditional UIs and intermediaries.

The move signifies a shift in how operational data is consumed. Increasingly, AI systems, not just human engineers, are tasked with monitoring and debugging complex systems. Tools like coding assistants can now investigate errors before a human even opens a ticket, and AI SREs can correlate failures across services without human intervention. This places new demands on observability platforms: they must be accessible not only via a graphical interface but also through programmatic means like APIs, MCP, and CLI.

The redesigned Observe by Snowflake MCP server and its accompanying CLI provide this programmatic access. Every function exposed by the MCP server to AI agents is now also available as a CLI command. This parity means engineers can build custom alert-triage agents that automatically query Observe for production telemetry when an alert fires, or develop copilots that assist human engineers during incident investigations. According to Snowflake's announcement, agents can now query Observe directly, enhancing speed and efficiency for reasoning about production environments.

Agent-Friendly Observability

The previous Observe CLI offered limited functionality. The new, agent-compatible CLI provides full programmatic access to the Observe platform's capabilities from the terminal. It supports agent environments like Claude Code and can be used in interactive terminal sessions. Some workflows can run autonomously in the background for routine tasks, while others are designed for interactive human-guided investigations. The CLI acts as a programmatic interface for composing, automating, and extending observability workflows.

Both the CLI and MCP server now include a library of prebuilt 'skills.' These are structured workflows based on common observability tasks, such as investigating incidents, tracing failures, validating changes, and detecting outliers. These skills are designed to work immediately after initial configuration, allowing both humans and machines to start using them without extensive setup.

Cost Efficiency and Direct Access

A key architectural change in the MCP server is the removal of a custom LLM intermediary. Previously, agents sent queries to a single endpoint where Snowflake's LLM interpreted the request. The new architecture allows agents to connect directly to Observe's APIs. This eliminates the cost and latency associated with the LLM middleman, making the platform more cost-efficient for both Snowflake and its customers. Direct API access also means agents can query Observe's context graph, identify the most valuable datasets, and write efficient OPAL queries for precise observability against any telemetry data.

This direct access extends to key platform capabilities, including APM service maps, OpenTelemetry data collection status, and active alert listings. These features are now directly queryable by agents through the MCP server, mirroring the functionality available in the Observe UI.

Snowflake positions this development as part of a broader trend in how observability data is consumed. They believe observability is fundamentally a data problem, and making this data directly accessible to AI agents and developer workflows is a critical step. For companies like Observe, which StartupHub.ai data shows has a score of 59/100 and has raised $156M (Series B, 2025) with a post-money valuation of $900M, this move could further differentiate them in a competitive market. Competitors such as New Relic and Dynatrace score 85/100, while Datadog is at 64/100 and Splunk, despite its historical market dominance, scores only 10/100, according to StartupHub.ai data. This strategic enhancement could boost Observe's appeal to organizations increasingly adopting AI-native engineering practices. We've previously explored how Cloudflare agents gained similar observability capabilities, highlighting a growing industry trend.

Engineers can access the new MCP server by logging into their tenant and navigating to "Manage account" → "MCP server details." Snowflake provides setup instructions and demo workflows for integrating agents.