Causely’s MCP server upgrades help background agents deploy faster, produce traceable decisions, and use fewer tokens, with new open source examples.
OpenTelemetry eBPF Instrumentation captures LLM calls, MCP tool calls, embeddings, and reranks at the network layer, giving multi-agent apps full traces with no code.
With the Causely MCP server, AWS DevOps Agent halved its investigation time and cut tool calls from 36 to 10. With an external fault, 3 minutes instead of 15.
In three regression scenarios, Claude Managed Agents with Causely produced correct fixes 3.6–5.7x faster and at 3.5–5x lower cost than the baseline agent.
A control band breach tells an autonomous agent only that a metric moved. A trigger carrying a causal chain tells it what broke and how far the problem reaches.
New Causely MCP tools show which diagnoses an agent ruled out and how far a failure would spread, so teams can verify the agent's reasoning before acting in production.
Operating AI agents in production takes three layers: instrumentation, evaluation, and causal reasoning, including the root-cause analysis behind agent reliability.
Causely joins the AAIF along with A2A from Google. Agents can now talk over an open standard, but what they say is only as good as the context they start from.
Open-Inspect gives background agents an execution environment. Causely gives them a causal diagnosis. Together, they close the loop from finding to fixing.
MCP tunnels let Claude Managed Agents reach and authenticate to real infrastructure, but access alone doesn't make an agent one you'd trust to run on-call.
On-call means re-triaging the same problem as diagnoses shift underneath you. Causely's new Issues give people and agents one stable thread instead.
Causely turns Dynatrace entities, topology, and alerts into a causal model that names the root cause, making on-call agents better, faster, and cheaper.