What is Cross-Tool Contamination?

1 min read Updated

A vulnerability where one MCP server's tool descriptions influence or override how agents use tools from other servers, enabling stealthy data exfiltration or privilege escalation across server boundaries.

WHY IT MATTERS

When an agent connects to multiple MCP servers, all tool descriptions enter the same context window. A malicious server can craft descriptions that instruct the agent to redirect data from trusted servers to attacker-controlled endpoints, or to use trusted tools in dangerous ways.

The agent can't distinguish between legitimate tool documentation and injected instructions — both are just text in its context. Per-server isolation at the policy layer prevents one server's tools from affecting another's.

HOW POLICYLAYER USES THIS

Intercept enforces policies per-server, ensuring that tool calls to one server can't be influenced by descriptions from another. Each server connection has its own policy scope.

FREQUENTLY ASKED QUESTIONS

How does this differ from tool poisoning?
Tool poisoning targets a single tool's description. Cross-tool contamination uses one server's descriptions to manipulate how agents interact with a completely different server.

FURTHER READING

Enforce policies on every tool call

Intercept is the open-source MCP proxy that enforces YAML policies on AI agent tool calls. No code changes needed.

npx -y @policylayer/intercept
github.com/policylayer/intercept →
// GET IN TOUCH

Have a question or want to learn more? Send us a message.

Message sent.

We'll get back to you soon.