Akuity gives Claude, Codex and Cursor a governed path to prod
From Argo CD's creators: an MCP scoping agents to your own permissions, with a policy gate. Also: OpenAI sunsets gpt-5.4-cyber, Nvidia ships an 84GB card.
Copy markdown
A badge for your agent, not a master key
Akuity — the team behind Argo CD and Kargo — shipped an MCP Server that lets Claude, Codex, or Cursor act on your Kubernetes pipelines authenticated as you, inheriting your exact permissions across clusters. A policy layer can block sensitive production actions even when you're cleared to run them, and every call lands in an audit trail, closing the raw-access, no-identity gap most MCP servers leave open.
What it actually hands the agent
The agent gets real operational context — deployment history, cluster health, and change lineage across the whole fleet — so it can bootstrap new services, promote releases under existing approval rules, and triage incidents. MLB's DevOps team says it surfaced more than 100 degraded apps within hours and traced a systemic issue to its root cause in under 10 minutes.
The catch: it's the platform, not a free drop-in
This rides Akuity's managed enterprise platform, and pricing and availability weren't disclosed — so it's aimed at teams already on managed Argo CD, not a weekend side project. The pattern worth copying regardless is identity-scoped MCP with a policy gate: the missing piece in nearly every agent-to-production setup.
Elsewhere: OpenAI sunsets gpt-5.4-cyber
If your code calls gpt-5.4-cyber, migrate to gpt-5.6-cyber before October 1, 2026 or the calls stop. OpenAI also retired ChatGPT's automatic Instant-to-Thinking switch, so you now pick reasoning mode by hand.
Elsewhere: Codex 0.154.0 adds reasoning controls
OpenAI's Codex CLI shipped 0.154.0 with finer-grained reasoning controls and ExternalMessage support for sync and async operations — more say over how hard the model thinks per step when you're wiring up agents.
Elsewhere: Nvidia's RTX Pro 5500 packs 84GB
Nvidia's new RTX Pro 5500 Blackwell workstation card carries 84GB of GDDR7 — the RTX 5090's 21,760-core count but 2.6x its VRAM, at a 600W limit. That's enough headroom to run far bigger models locally without splitting them across cards.