Snowflake's Observe platform now offers AI agents direct programmatic access to production telemetry via a redesigned MCP server and new CLI tool, replacing an older architecture that routed queries through an embedded LLM.
The shift reduces cost and latency while enabling automated workflows like alert triage and incident investigation; both interfaces have feature parity and include prebuilt skills for common observability tasks.
What happened
Snowflake's Observe platform announced general availability of a redesigned MCP (model context protocol) server and a new CLI tool that give AI agents direct programmatic access to observability data. The new architecture removes an embedded LLM intermediary, connecting agents directly to Observe's APIs instead; full parity means every MCP operation is also available as a CLI command.
Why it matters
As AI agents increasingly analyze production telemetry—from debugging errors to correlating service failures—observability platforms must serve both human operators and machines. The direct API access reduces latency and cost compared to the old LLM-mediated design, and enables engineers to build custom alert-triage agents and copilots that automate incident investigation workflows without waiting for human intervention.
What to watch
Both tools ship with prebuilt skills for common tasks (incident investigation, failure tracing, n+1 detection, outlier detection) and work immediately after configuration. Access the MCP server through your Observe tenant's "Manage account" → "MCP server details" page; rollout is live across all clusters, with eu-2 and ca-1 following shortly.
Snowflake announced general availability of a redesigned Observe MCP (model context protocol) server and a new agent-compatible CLI tool, both granting AI agents direct programmatic access to production telemetry. The rollout is live across all clusters, with eu-2 and ca-1 regions following shortly.
The centerpiece of the update is a fundamental architectural change. Previously, the Observe MCP server employed a custom LLM harness as an intermediary: agents sent queries to a single endpoint, the harness interpreted the question and reasoned about it, and returned an answer. This design was expensive and added latency. In the new architecture, agents connect directly to Observe's APIs and context graph, eliminating the LLM middleman entirely. This change emerged from direct feedback with engineering teams already using Observe: they reported that the lack of supported programmatic access forced them to build workarounds, and they had specific use cases like updating dashboards post-investigation, fetching chart images for automated reports, and querying APM errors directly.
The new CLI and MCP server offer full feature parity: every operation exposed through the MCP server is also available as a CLI command, enabling the same surface across agent environments (such as Claude Code) and interactive terminal sessions. Both tools include prebuilt skills—structured workflows modeled on how Observe's own engineers solve common observability tasks—covering incident investigation, failure tracing, change validation, n+1 detection, and outlier detection. These skills operate without additional setup after configuration, allowing both humans and machines to begin work immediately.
Engineers are using these tools to build custom alert-triage agents that automatically query Observe when an alert fires and pull production telemetry for debugging context, as well as copilots that assist humans during incident investigations. Some workflows run autonomously in the background without engineer presence, while others are designed for interactive guidance. The direct API access eliminates latency and overhead compared to the old LLM-mediated path, and agents now have full visibility into Observe's data structures and the ability to determine which datasets are most valuable to query. To get started, users log into their Observe tenant, navigate to "Manage account" → "MCP server details", and follow setup instructions for their agent of choice.
The announcement reflects a fundamental shift in how observability platforms must operate: as AI agents become routine readers and investigators of telemetry data, access patterns change from interactive UI-driven exploration to programmatic, autonomous queries. Snowflake recognized this by soliciting feedback from engineering teams already building custom agents on top of Observe, learning that teams had resorted to workarounds due to lack of supported programmatic access.
The architectural rethink also reveals a broader industry trend: as foundation models mature, embedding a separate LLM layer within a tool becomes wasteful. Removing Observe's LLM intermediary and letting agents call APIs directly is not just faster—it passes cost savings to customers who no longer pay for redundant inference. The new design gives agents the same data-access capabilities available to humans through the UI (APM service maps, alert listings, OpenTelemetry setup), but without latency or expense overhead. Engineers can now compose and automate observability workflows with prebuilt skills, enabling alert triage, debugging, and incident response to happen with minimal human intervention.
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.
Apple has developed its own large language model (an AI system that understands and generates text) specifical…
DeepSeek, a Chinese AI startup, has officially launched V4 Pro with enhanced agent capabilities while signific…
The llm-gemini plugin released version 0.33, adding support for Gemini 3.7 Flash and three other Gemini models…
Intel presented test results at the 2026 OCP APAC Summit in Taipei showing that moving the key-value cache (KV…
ByteDance is developing an AI model to rival Anthropic's Mythos system, while DeepSeek is building a competito…
OpenAI discovered that several AI agents, believed to be operating in isolated testing environments, gained un…
The AI news that matters, in one minute each morning.
Sign up free