We scanned over 7,500 MCP servers. 9.2% have critical vulnerabilities. 36.7% have SSRF exposure. 43% have command injection flaws. Before connecting to an MCP server, check: https://t.co/3jroP4AtaN

Gateways are not flawed. They’re operating at the correct layer for what they were designed to do. But agentic systems have a second layer — and that’s where the failures emerge. #AgenticObservability

The Agentic Execution Gap: the space between an AI agent’s decision and its outcome, where gateways, prompt tracing, and traditional observability have zero visibility. That’s where incidents actually happen. #MCP #AgenticAI

The next major AI agent incident won’t involve bypassing authentication or exploiting a misconfigured policy. It will involve approved access, valid tool usage, clean gateway logs — and unexpected side effects. #AgenticObservability #AIAgents

Same Sandbox. Same observability: MCP events, process execution, file access, network connections. New: we host it for you. Plus secure runtime and FastMCP pre-deployed. Register and get your first agent trace the same day. https://t.co/7aP59gINCn #AIAgents #MCP

We kept hearing: "I want to see what my agent does, but I need to set up AWS first." So we removed that step. Managed PaaS for the Agentic Observability Sandbox — same full observability, zero infrastructure, free 30-day trial. https://t.co/7aP59gIfMP #AgenticObservability

The Agentic Observability Sandbox now has a managed PaaS option — free. Register → credentials same day → start tracing agent execution. No AWS setup required. https://t.co/7aP59gINCn #AgenticObservability #MCP #AIAgents

Harold Byun (CPO) and David Onwukwe (Principal SE) will show you: → Where MCP gateway visibility ends → Why developers work around gateways — and why that creates risk → What end-to-end execution visibility actually means → Live demo: execution-level control in action

https://t.co/BW9hYJBATZ

[Live webinar] What MCP Gateways Miss: Observing and Controlling Agent Execution February 26, 2026 | 11:00 AM PST | 30 min Gateways control requests. Agents operate at the execution layer. That's a different problem. Register 👇

The 5 MCP challenges that are preventing your agents to pivot from POCs to prod.

Agents aren't scary. Unobservable agents are.

Shadow MCP is the new shadow SaaS—except it can take actions, not just store data. Devs connect agents to unvetted servers. Gateways can't see what executes inside tools. Most teams have zero post-incident forensics for agent actions.

Full details →

56% of enterprises have AI agents in production. Most can't answer: "what did the agent actually do?" Tool call ≠ tool execution. That gap is where incidents live. 5 MCP challenges every team hits past the demo stage. More details in 🔗👇

Jason Meller's full post:

We built the MCP Trust Registry to fix this. Free security ratings for any MCP server. Check before you connect. Free MCP server ratings:

We've scanned 8,200+ MCP servers. What we found: - 9.2% have critical vulnerabilities - Auth gaps, SSRF exposure, path traversal - Most have no publisher verification - 20,000+ new servers published every month Most teams let agents connect with zero security diligence.

The #1 downloaded @openclaw skill was malware. @1Password's latest blog broke down how agent "skills" became a full malware delivery chain. The conclusion: agent registries are the next supply chain attack surface. The data backs it up. 🧵

Our research team found this vulnerability using the MCP Trust Registry. The bigger story: 36.7% of MCP servers we scanned have potential SSRF exposure. Check your servers before you connect:

Microsoft & Anthropic MCP Servers At Risk of RCE, Cloud Takeovers: https://t.co/IIhoSLHeQ1 by Nate Nelson