Snowflake has made its Observe observability platform accessible to AI agents through a redesigned MCP server and new CLI, both now in general availability.
The key change is architectural: agents now connect directly to Observe's APIs rather than querying through an LLM middleman, reducing latency and cost while giving agents the same visibility into observability data that humans have through the UI.
Engineers are already building custom alert-triage agents and incident investigation copilots with these tools.
What happened
Snowflake announced general availability of a redesigned Observe MCP (model context protocol) server and a new Observe CLI with full parity, allowing AI agents and engineers to query production telemetry directly through APIs, MCP, and command-line interfaces. The new MCP server connects agents directly to Observe's APIs instead of routing through an LLM intermediary.
Why it matters
As AI agents increasingly handle observability tasks—such as automatic alert triage, incident investigation, and error debugging—observability platforms must serve both humans and machines. The redesigned MCP server removes latency and cost by eliminating the previous LLM intermediary, enabling agents to reason about production data faster and more efficiently without passing through a reasoning layer.
What to watch
Both tools ship with prebuilt skills for common observability workflows (incident investigation, failure tracing, change validation, n+1 issue detection, outlier detection) and work immediately after configuration. General availability is rolling out now across all clusters; eu-2 and ca-1 regions will follow shortly.
Snowflake announced the general availability of a redesigned Observe MCP server and a new Observe CLI, both designed to allow AI agents and engineers to access and query production telemetry directly. The announcement reflects a broader shift in observability: as AI agents become sophisticated enough to investigate errors, correlate failures across services, and automate incident triage, observability platforms must serve machines as well as humans.
The previous Observe CLI covered only a narrow slice of platform functionality, with most capabilities inaccessible from the terminal. The new CLI provides programmatic access to the entire Observe platform from the command line, matching the capabilities exposed through the MCP server and the web UI. Both tools work in agent environments like Claude Code as well as interactive terminal sessions. Engineers use these tools to build custom alert-triage agents that automatically query Observe when an alert fires and pull production telemetry for debugging context, and copilots that assist during incident investigations.
The most significant change is architectural. The original MCP server included a custom LLM harness that received user queries, performed reasoning, and returned answers. This design created latency and cost overhead. The new architecture removes the LLM intermediary entirely. Agents now connect directly to Observe's APIs—the same APIs powering the frontend—giving them direct visibility into Observe's data structure and the ability to query the context graph, determine which data sets are most valuable, and write efficient OPAL queries. This removes both the latency of passing through a reasoning layer and the cost of operating an LLM for every agent query.
Both the CLI and MCP server ship with prebuilt skills: structured workflows based on how Observe engineers solve common observability tasks. These skills enable agents and engineers to investigate incidents, trace failures, validate changes, find n+1 issues, and detect outliers immediately after configuration, without additional setup. The new architecture gives agents access to the same capabilities available in the Observe UI: APM service maps, OpenTelemetry data collection setup, and active alert listing. General availability is rolling out now across all clusters, with eu-2 and ca-1 regions following shortly. Engineers can get started by logging into their Observe tenant, navigating to "Manage account" → "MCP server details", and following setup instructions for their agent platform.
Observability has traditionally been a human-centric practice: engineers read dashboards, logs, and metrics to understand system behavior. As AI agents become capable of reasoning about operational data, observability platforms face a new demand: their data and capabilities must be accessible not just through graphical interfaces but through programmatic endpoints that agents can call autonomously. Snowflake's redesigned Observe MCP server and CLI represent a direct response to this shift.
The architectural change is significant: the original MCP server included a built-in LLM that received queries from agents, performed reasoning, and returned answers. This design incurred latency and cost overhead. The new server eliminates that intermediary, giving agents direct access to Observe's APIs—the same ones powering the UI. This means agents can explore the context graph, determine which data sets are most valuable, and write efficient OPAL queries without waiting for a reasoning layer to interpret their intent. The feedback from engineering teams already using agents with Observe shaped this redesign; several teams had built workarounds because programmatic access was limited, and specific use cases (updating dashboards, fetching chart images for reports, querying APM errors) required new platform capabilities.
The CLI achieving full parity with the MCP server ensures that the same operations available to agents are available to engineers working from the command line. Prebuilt skills for incident investigation, failure tracing, and outlier detection ship with both tools, allowing both humans and machines to use them immediately after configuration. This represents a foundational shift in how observability is consumed: not as a product for human inspection, but as a data platform accessible to the full stack of AI and developer tools a team relies on.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
Ask AI anything about this article. Q&As are published on this page for other readers too.
Morgan Stanley analysts assessed how lower-cost open-weight AI models—which users can download and run on thei…
Apple has formally launched its 'Ads on Maps' platform, allowing businesses to purchase promoted placements in…
Brookfield Corporation has completed acquisitions that expand its life and annuity insurance operations and an…
Alibaba Group's Qwen family of open-weight AI models accumulated more than 3 billion global downloads in the p…
Snowflake announced general availability of a redesigned Observe MCP server and a new Observe CLI with full pa…
Parnassus Core Equity Fund sold its Salesforce holding in Q2 2026, citing declining conviction in horizontal a…
The AI news that matters, in one minute each morning.
Sign up free