PolicyLayer

PolicyLayer provides a control gateway for AI agents interacting with Model Context Protocol (MCP) servers, enabling oversight of every tool call before it is executed. The platform is designed to allow, deny, require approval, hide, or rate-limit each MCP tool call based on deterministic policies, ensuring that actions taken by AI agents are governed by explicit, enforceable rules rather than relying on system prompts within models. This approach addresses the challenge of prompt injection and the inability of models to enforce their own operational boundaries, as PolicyLayer sits outside the model and evaluates intent at the point where an action is about to be taken.

The service operates without requiring an SDK or additional infrastructure deployment, and it supports live policy changes. Users can connect existing MCP servers or HTTP endpoints and use proxy URLs and grant tokens generated by PolicyLayer to govern real agent traffic. The platform is compatible with various MCP clients, including Claude, Code Cursor, Codex, GitHub Copilot, VS Code, and Gemini, and can integrate with services such as Stripe, GitHub, and AWS. PolicyLayer's workflow involves identifying the caller (person, agent, environment, or workflow), evaluating the intended action and its arguments, and recording a verdict that includes the grant, policy version, matching rule, and upstream result. Argument values are excluded from ordinary logs for privacy, while keys and matching rules are retained.

PolicyLayer supports issuing scoped grants to both people and agents, separating the roles of actors and decision-makers. It enables holding actions for human approval when necessary, and every action is logged with an immutable, append-only audit record. The system enforces a fail-closed posture, denying actions when grant, policy, or upstream states are ambiguous. Credentials are encrypted at rest using AES-256-GCM and are only decrypted when needed to call the upstream server; they cannot be retrieved through the dashboard or API after entry. The platform also provides a registry that continuously classifies public MCP servers and their tools, offering policy intelligence that users can adapt to their own workflows.

The tool is suited for organizations deploying AI agents in production who require granular, auditable, and enforceable control over automated actions taken via MCP tool calls. It emphasizes security, traceability, and compliance by ensuring that nothing is executed without a policy-backed verdict and that every decision is attributable and immutable.