PolicyGuard

PolicyGuard is a Python MCP server that enforces security and governance for AI agents by validating actions, tracking incidents, and monitoring compliance. It sits between agents and their tools, ensuring only permitted operations are performed and providing an auditable trail for governance.

How to use

You use PolicyGuard by connecting an MCP client to its MCP API and invoking its tools to validate actions, register agents, create policies, and view audits and compliance status.

How to install

Prerequisites you need before installing PolicyGuard:

• Python 3.10+ is required for the MCP server runtime.

• Use pip to install Python package dependencies.

Step-by-step installation commands you should run in your shell:

Configuration and usage details

PolicyGuard provides six MCP tools for governance and security. You will typically enable an MCP HTTP connection in your environment or run the server locally in HTTP mode to test or develop.

Security model and governance

Trust levels define how your agents are evaluated before performing actions. The model includes a layered evaluation order to ensure only permitted operations proceed, with an admin level granting full access and lower levels enforcing stricter checks.

Project structure and running locally

To run PolicyGuard locally, start the server in HTTP mode or run it directly as a local process.

Available tools

validate_action

Primary tool to check if a requested action is allowed before the agent executes it.

register_agent

Register a new agent with its identity and trust level for governance.

create_policy

Create security rules that govern what actions are permitted or denied.

get_audit_log

Query the audit trail of agent actions for review and compliance.

get_compliance_status

Retrieve the security dashboard metrics to assess overall posture.

report_incident

Manually report security incidents to track and respond to events.