Skip to content
StrikeCyberStrikeCyber
Field Manual · Attack technique

MCP Tool Poisoning

Reviewed November 13, 2026Attack technique 1 min read

Quick answer

MCP tool poisoning is an attack on AI agents that use the Model Context Protocol (MCP). A malicious or compromised MCP server hides instructions in its tool descriptions or responses. The agent reads them as trusted context and can be manipulated into leaking data or misusing other connected tools.

On this page
01

How It Works

MCP lets an AI agent discover and call tools provided by servers. Each tool comes with a description that the model reads to decide how to use it. An attacker who controls a server, or publishes a useful-looking one, can put instructions in that description or in tool results. Because the model treats them as part of its context, those instructions can redirect the agent, including how it uses tools from other, legitimate servers.

Variants include servers that behave safely at first and change their descriptions after installation, and tool names that shadow trusted tools.

02

Why It Matters

Agents increasingly hold real permissions: file access, email, code repositories, ticketing and cloud consoles. A poisoned tool can turn that access against the organization without any user noticing. See Model Context Protocol security.

03

How to Test for It

  1. Inventory every MCP server and tool each agent can reach, including where each came from and who maintains it.
  2. Review tool descriptions and schemas for hidden or excessive instructions.
  3. Introduce a controlled malicious server in a test environment and check whether it can influence other tools.
  4. Test whether tool descriptions can change after approval without review.
  5. Measure what a hijacked agent could do with its current permissions.
04

How to Reduce the Risk

Allow only vetted MCP servers, pin and review tool definitions, isolate high-privilege tools, require confirmation for sensitive actions, and log tool calls with the context that triggered them.

05

Regulatory and Framework Context

For US organizations adopting AI agents, MCP reviews fit naturally alongside existing third-party and supply chain risk processes, and support NIST AI RMF and ISO/IEC 42001 controls on AI system suppliers.

Latest intel on this topic

Updated hourly from the daily briefing.

Common questions

What is the Model Context Protocol?

An open protocol that lets AI applications connect to external tools and data sources through a standard interface. Each MCP server exposes tools an agent can call.

Are official MCP servers safe?

Safer, but not automatically safe. Any server's tools extend what an agent can do, so permissions and descriptions still need review.

Change log

  1. November 13, 2026First published.
(877) 657-8496