531 posts tagged with "AI agents"
AI agents and autonomous systems
MCP's Stateless Core Just Shipped: What a Deploy-From-Chat MCP Server Gets to Delete
MCP dropped protocol-level sessions on July 28, 2026 — deleting the sticky routing and shared session store a self-hosted deploy/rollback MCP server used to need, and adding a formal lifecycle policy worth pinning to.
MCP's Tasks Extension Kills the Deploy-and-Poll Hack — But Idempotency Is Still Your Job
MCP's new Tasks extension finally standardizes the poll-for-status hack every deploy MCP server reinvents — but idempotency, cancellation, and state ownership are gaps the spec leaves for the platform to close itself.
The Mexico Breach Wasn't a Jailbreak. It Was a File Claude Reloaded on Every Run
A single attacker jailbroke Claude Code once, then breached nine Mexican government agencies across 34 sessions without ever repeating the jailbreak — because Claude wrote its own jailbroken instructions into a file it auto-loads every run. Here's the exact mechanism, and what it means for any agent-operated deploy pipeline that trusts persisted state.
OpenAI Adopted Anthropic's Agent Skills Format in 48 Hours — What That Means for Your Deploy Runbooks
OpenAI adopted Anthropic's Agent Skills packaging format within 48 hours of its release, not just its MCP protocol. Here's what that convergence changes about writing a portable runbook Skill instead of a bespoke one.
Vercel Shipped an Official MCP Server. Here's Every Tool It Refuses to Give Your Agent
Vercel's official MCP server took a year to gain its first write tool — and it still can't touch an existing project's env vars, domains, or a rollback. A tool-by-tool audit of the agent-ops bar the market leader actually set.
The AI Rollback Paradox: 81% of the Best-Governed Teams Pulled Their Agents Anyway
Sinch's 2026 survey found 74% of enterprises rolled back a live AI agent — and the best-governed teams rolled back the most, at 81%. Here's what that means for building rollback into an MCP deploy surface as a cheap default instead of another approval gate.
Your Deploy Agent's Sandbox Will Die Mid-Rollout — Here's How Its MCP Server Should Survive That
MCP's July 2026 spec update made the protocol stateless — which means the burden of surviving a crashed sandbox now falls entirely on the MCP server. Here's the Postgres checkpoint design, borrowed from LangGraph, that a deploy-from-chat PaaS needs to make a multi-step rollout outlive the agent that started it.
Designing upgrade_cluster: An MCP Tool Contract for Fleet-Wide Kubernetes Node Upgrades
A deploy/rollback MCP tool that goes wrong costs a bad release. An upgrade_cluster tool that goes wrong can leave a fleet straddling a Kubernetes version-skew boundary mid-drain. Here's a concrete tool contract — plan mode, per-node health gates, and a hard stop — for handing an agent that authority safely.
The Rollback Floor: Designing a Deploy-From-Chat MCP Server That Can't Roll Back Too Far
A dashboard's rollback button only ever offers 'go back one.' An MCP tool wrapping the same API can't assume that restraint — here's a concrete tool contract, server-side floor, and audit schema that stop an agent from rolling back past the last known-good release.