Assay

Adoption judgment: Assay looks promising as a narrowly focused, early security component for MCP agents, but the available evidence is not enough to call it production-proven or broadly adopted. The evidence set contains one GitHub repository page and one X post. That is enough to establish the project’s stated direction and some attention, but not enough to verify operational maturity, compatibility across environments, or real deployment scale.

The repository describes a policy-as-code approach that can deny risky tool calls before they run and provide verifiable evidence of what actually ran. It also describes egress enforcement through eBPF/LSM on Linux, with a deterministic and offline positioning. A more accurate analogy is a pre-execution policy gate for agent tool calls combined with a Linux-level egress enforcement layer. It should not be confused with an MCP server, an agent runtime, a general API gateway, or a standalone logging and prompt-guarding product.

The likely barriers are writing and maintaining policies, connecting the control point to MCP tool calls, and operating Linux eBPF/LSM components.