Deep dives, practical guides, and incident analyses from engineers who build Safeguard. No fluff, no vendor FUD — just what you need to ship secure software.
Tier an MCP server by what it can do: read-only public data is low, writes to internal systems are medium, shell, secrets or production access are high.
A works with MCP badge says little. Ask which role, which transport, which authorization and which tools, then check each against the MCP specification.
Prompt injection spans direct and indirect attacks, encoding tricks, repeated and multi-turn attempts, markdown leaks, RAG poisoning and agent abuse.
OWASP LLM08 covers flaws in how RAG systems store and retrieve embeddings: data leakage, tenant mixing, inversion and poisoning. Start with access control.
PKCE stops a stolen OAuth authorization code from being redeemed by an attacker. The MCP spec requires S256 and says to refuse servers without support.
ONNX can store tensors in separate files named by a relative path. Learn how external data works, the checksum field, and why path handling is a risk.
MCP sampling lets a server ask the client for a model completion. Learn the risk, the human review the spec expects, and why the draft marks it deprecated.
A safetensors file is an 8 byte length, a JSON header, then raw tensor bytes. See the layout, the checks built in, and what the format cannot promise.
Excessive agency means an AI system can take damaging actions it should not. Learn the three causes, MCP examples of each and the OWASP controls that help.
Weekly insights on software supply chain security, delivered to your inbox.
Your first fix PR is minutes away.
No sales call required, even your agent can complete the purchase over MCP.
