Publishing a security-critical component as open-source MIT code is a deliberate architectural choice, not a marketing gesture. If the enforcement layer that sits between your AI agent and every tool it calls is a black box, you cannot audit it, you cannot fork it, and you cannot trust the compliance guarantees it claims to produce.
That is why Vaikora Guard MCP is published on PyPI as `vaikora-guard-mcp`, written in Python, MIT-licensed, and self-hostable. This post explains what the license means operationally, how the enforcement model works, and what stays on your infrastructure versus what the commercial control plane adds.
What "MIT-licensed and self-hostable" means in practice
MIT is a permissive license. You can read the source, modify it, run it inside your own infrastructure perimeter, and never send a request to Data443’s servers. For security teams who need to control every hop a request makes, that matters.
Self-hosting means the enforcement proxy runs in your environment. No agent action payload leaves your perimeter before the policy decision is made. The gateway inspects the action, evaluates it against your policies, and returns allow, block, modify, or require-approval before the action executes. The LLM provider call or tool call only proceeds if the policy says yes.
For regulated environments, HIPAA, PCI-DSS, and GDPR compliance presets ship with the package, so you are not writing policy from scratch.
How the enforcement model works
Vaikora Guard MCP sits inline between your AI agent and the LLM or tools it calls. Every action passes through six content inspection modules before execution. The decisions are deterministic: identical input always produces an identical output. That property matters for audit because it means every outcome is explainable.
A concrete scenario: an AI agent is tasked with summarizing customer support tickets and has access to a database query tool. Without a policy gate, the agent can construct and run any query the underlying database role permits. With Vaikora Guard MCP inline, every query action is evaluated before it executes. A policy rule that blocks SELECT statements touching a table containing payment card data stops the query at the gate, writes the decision to the audit log, and returns a block response to the agent. The agent never reaches the database.
That decision, along with every other decision in the session, is signed into a SHA-256 hash-chained, tamper-evident audit log. The chain gives you an ordered, verifiable record of what was allowed and what was blocked.
Enforcement latency is P50 8ms and P99 45ms, so the gate does not introduce meaningful overhead for most agent workflows.
The MCP compatibility and what it covers
MCP (Model Context Protocol) is a standard for how AI agents describe and invoke tools. Vaikora Guard MCP is compatible with MCP-based agent architectures, which means it can sit in front of any MCP server without requiring you to rewrite your agent code.
The gateway also ships adapters for 12 LLM providers, so if your agent calls a hosted model API directly rather than through MCP, the same enforcement applies. No SDK migration is required.
What the open-source gateway does not include
The gateway enforces policy. It does not give you a central place to write, version, distribute, and audit policies across multiple agents running in multiple environments. That is the Vaikora Control Plane, the commercial product hosted by Data443.
The control plane adds:
- A policy editor and version history so policy changes are tracked
- Centralized audit log review across all agent deployments
- Native integrations with Microsoft Sentinel, Microsoft Defender, SentinelOne, CrowdStrike Falcon, and AWS Security Hub
- Threat Intelligence indicator push into SentinelOne with severity mapping (Critical=100, High=75, Medium=50, Low=25)
- Promotion of high-severity AI agent risk into Azure Security Center and Microsoft Defender for Cloud as native incidents
If you are running one agent in one environment for internal testing, the open-source gateway alone may be sufficient. If you are running agents in production across multiple teams and need to demonstrate compliance, the control plane is where policy governance becomes auditable at scale.
How to decide what you need before you install anything
Before installing, answer three questions:
- Do you need policy decisions to be auditable by an external auditor or regulator? If yes, you need the hash-chained log, which the open-source gateway provides.
- Do you need policies to be centrally managed and versioned across more than one agent deployment? If yes, the control plane is necessary.
- Do you need AI agent threats to surface automatically in your existing SIEM or EDR? If yes, the control plane’s integrations handle that without custom engineering.
The open-source gateway is the right starting point regardless. Install `vaikora-guard-mcp` from PyPI, review the inspection module configuration, and run a session with a test agent to see how the audit log entries are structured. That gives you a concrete picture of what policy enforcement at the action level actually produces before you decide how to scale it.