AI Signal 539
Show HN: Product analytics (and evals) for agent sessions on your MCP
↗
For teams shipping an MCP server, the standard logs only show tool calls; the user's goal and the agent's reasoning around those calls live inside the client and are otherwise unobservable. Armature positions itself as the product-analytics layer for that gap, and the differentiator it claims over LangSmith and Langfuse is a product-team focus on user outcomes rather than engineer-facing agent observability. Only one feed has carried the launch, so the offering is uncorroborated elsewhere.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
Integration is an SDK wrap around the server you already run, available in TypeScript, Python, and Go, pitched as requiring only a few lines of code with no change to existing MCP behavior.
The service reconstructs user intent, the agent's intermediate reasoning, and the full tool call trace, then scores each session on success, groups sessions into use cases, and clusters failure modes even when every API response returned 200.
PII and secrets are redacted by detection models before storage, the free tier covers 1,000 sessions per month with usage-based pricing beyond that, but only a single feed has surfaced the launch so the claims carry no independent confirmation.
THE CLUSTER