mcp-security
Safeguard articles tagged "mcp-security" — guides, analysis, and best practices for software supply chain and application security.
59 articles
LLM Insecure Plugin Design Vulnerabilities
ChatGPT plugins, LangChain agents, and MCP servers have all shipped insecure plugin flaws exposing accounts and data. Here's how Safeguard defends against them.
Excessive Agency in LLM-Powered Applications
Excessive agency turns a bad LLM output into an executed action. From Replit's July 2025 database deletion to Air Canada's chatbot ruling, here's what it is and how to scope it down.
Prompt Injection Attack Techniques and Defenses
Prompt injection is now OWASP's #1 LLM risk, and real incidents like EchoLeak and Slack AI prove it can mean zero-click data exfiltration. Here's how it works and what stops it.
The postmark-mcp Backdoor: What MCP Server Vetting Should Look Like
A trojanized MCP server BCC'd every email it sent to an attacker for weeks, downloaded 1,643 times, before anyone noticed. Here's the pattern and the fix.
Least-privilege scoping for AI agents with write access to code, CI, and cloud
OWASP's 2025 LLM Top 10 names Excessive Agency a top risk; a single over-scoped CI token already dumped secrets from 23,000+ repos in 2025.
Model Context Protocol in 2026: News, Adoption, and Security Landscape
The latest Model Context Protocol news: foundation governance, near-universal vendor adoption, a maturing spec, and a security track record that already includes critical CVEs.
A Reproducible Rubric for Measuring Prompt-Injection Risk in Agent Skills
OWASP has ranked prompt injection the #1 LLM risk for two straight editions, yet almost no one scores agent skill packages for it consistently. Here's a rubric.
Governing AI agents inside the execution loop
Snyk's Evo Agentic Development Security, in open preview since June 23, 2026, hooks directly into an agent's tool calls — proof that pre-deployment review can't govern a decision made mid-session.
How to build and justify an AI security budget
CVE-2025-6514 let a flawed MCP proxy escalate to full remote code execution — a preview of why AI/agentic risk needs its own budget line, not a slice of the AppSec line.
New security risks across the agentic development lifecycle
Snyk found 76 confirmed-malicious skills among 3,984 analyzed and roughly a third of public MCP servers carrying exploitable flaws — legacy AppSec never modeled an agent as the author.
Protestware via prompt injection: when maintainers target AI agents
jqwik 1.10.0 shipped a hidden instruction telling AI coding agents to delete their own tests, then erased it from the terminal with ANSI codes — protestware built for agents, not humans.
Securing MCP integrations for enterprise AI assistants
Claude's May 2026 enterprise and desktop expansion put MCP tool calls in front of compliance teams and IDEs alike — governance can no longer be an afterthought.
Self-healing security runs on Safeguard.
Your first fix PR is minutes away.
No sales call required, even your agent can complete the purchase over MCP.