Why does installing one MCP server let it silently poison every tool in my pipeline?
Opportunity
MCP has become the standard way to extend what AI agents can do, with hundreds of community-built servers distributed through public registries. These servers run with the agent's full permissions and can embed hidden instructions inside their tool description strings, instructions the model treats as authoritative context rather than user input. When researchers scanned 1,899 open-source MCP servers they found 5.5 percent contained description patterns consistent with tool poisoning, and a poisoned server's description is injected into the shared context window the moment it is installed, contaminating behavior toward entirely unrelated tools with no invocation required. There is no signing standard, no sandboxing, and no hash-pinning mechanism for MCP servers, and a systemic flaw disclosed in April 2026 affects an estimated 200,000 instances across more than 150 million package downloa
Why it matters
The MCP registry is growing at the speed npm once grew and embedding the same supply-chain vulnerabilities at the protocol layer rather than fixing them before conventions harden.
How I score the opportunity
The Opportunity Score is my own read, not a measurement: how much it hurts, how often it bites, and how little exists to solve it today. Higher means I think it is more worth building.
How much pain it causes when it shows up.
How often people actually run into it.
How little good tooling exists for it today.
More problems worth solving
Why does every AI app forget me the moment I close the tab?
AIWhy is learning a new field still gated by knowing what to ask?
AIWhy can a non-expert not verify what an AI just told them?
AIWhy do we test models on benchmarks but ship them on vibes?
AIWhy do AI agents have no memory of their own mistakes?
AIWhy can't I audit what a model was actually trained on?