MAQPNA

Learn

Is MCP safe? MCP security, risk by risk.

The protocol is not the problem. What you connect, with which credentials, and who checks each call, is.

Updated 1 November 2026.

FAQ

Common questions

Is MCP safe to use?

Yes, if you control which servers agents can reach, give them short-lived scoped credentials, check each call against policy and keep a record. The protocol does not do these things for you.

Are local MCP servers safer than remote ones?

Not automatically. A local server runs with your privileges on your machine. The MCP security best practices ask clients to show the exact command before running one and to sandbox it with minimal file system and network access.

What is MCP tool poisoning?

An attack where a tool's description contains hidden instructions that steer the model, or where a server changes a tool after you approved it. Pinning tool definitions by hash and reacting to changes is the main defence.

Does OAuth make MCP secure?

OAuth authenticates the client and limits scopes. It does not decide whether a particular call with particular arguments is safe, and it does not stop data leaving. You still need per-call policy, DLP and audit.

Is there an OWASP list for MCP?

Yes. OWASP's MCP Top 10 covers risks such as token mismanagement, scope creep, tool poisoning, supply chain attacks and lack of audit. It was in beta as of 1 November 2026.

Put a decision in front of every MCP call.

Run MAQPNA locally in one command and watch each tool call get allowed, held or denied.

More install options
curl -fsSL https://maqpna.com/install.sh | sh