Tool poisoning
Tool poisoning is an attack where a malicious tool, often an MCP server, hides instructions in its tool description or metadata; the AI model reads them as guidance and may leak data or misuse other tools without the user noticing.
Also called: MCP rug pull, tool poisoning attack, TPA, tool shadowing
By Divya Sharma · Checked Nov 15, 2026 · How we review
When an agent connects to an MCP server, the server tells it which tools exist and describes what each one does. The model reads those descriptions in full to decide how to call them. The user usually sees only a short tool name. That gap is what tool poisoning exploits.
How it works
Invariant Labs, who named the attack in 2025, describe it as malicious instructions embedded in MCP tool descriptions that are invisible to users but visible to AI models. A tool that claims to add two numbers can carry hidden text telling the model to read local configuration files or keys and pass them along in an innocent-looking argument. Invariant describe it as a specialised form of indirect prompt injection, and its usual goal is data exfiltration.
Rug pulls and shadowing
An MCP rug pull is when a server you already approved changes its tool descriptions later, so a clean first review proves little. Tool shadowing is when a malicious server's description changes how the agent uses a different, trusted server's tools, for example redirecting where emails are sent, so the harmful tool never needs to be called itself.
Defences
Invariant suggest showing full tool descriptions to users, pinning server versions with a hash so changes are detected, and stricter boundaries between servers. Other common steps: install servers only from publishers you trust, as with any AI supply-chain risk; keep agent permissions narrow; and scan servers with tools such as Snyk Agent Scan. Each MCP tool review on this site lists the permissions a server asks for.
Related terms
- Prompt injection
- Prompt injection is an attack where text fed to a language model, typed by a user or hidden in a web page, file or tool result, changes the model's behaviour in ways its developer did not intend.
- MCP server
- An MCP server is a program that exposes tools, data resources or prompt templates to AI applications through the Model Context Protocol, so any MCP-compatible client can use them without custom integration code.
- AI supply-chain attack
- An AI supply-chain attack compromises an AI application through something it depends on, such as a downloaded model, a LoRA adapter, a dataset, a Python package, an MCP server or an agent skill, rather than attacking the application directly.
- Agent permissions
- Agent permissions are the rules that decide which tools, files, commands, networks and accounts an AI agent may use, and which actions need a human's approval first, so a mistaken or manipulated agent can do limited damage.
- Data exfiltration (AI agents)
- Data exfiltration, for AI agents, is when an attacker tricks an agent into reading private data it can access and sending it out, through a web request, an email, an image link or another channel the attacker controls.