Vaikora › Blog › Product Updates

Open-sourcing Vaikora Guard MCP: what the MIT license means for teams that need auditable AI agent control

Product Updates · Nov 13, 2026 · 4 min read

The MIT license on vaikora-guard-mcp is not a marketing gesture. It means your security team can read exactly what happens between an AI agent's tool call request and execution, verify the enforcement logic, and run it in your own infrastructure without a commercial dependency.

That specificity matters because AI agent tool calls are where policy violations actually happen. Not in the model, not in the UI: in the moment an agent issues a call to a database, an API, a file store, or an external service.

What the proxy actually does

vaikora-guard-mcp is a Model Context Protocol compatible proxy, written in Python and published on PyPI under the package name vaikora-guard-mcp. It sits inline between an AI agent and the tools or LLMs it calls. Before any action executes, the proxy evaluates the request against policy and returns one of four deterministic decisions: allow, block, modify, or require-approval.

Deterministic here means something specific. Identical input produces identical output, every time. There is no probabilistic scoring that lets a borderline request through on Tuesday because a threshold shifted. The decision is the same at P50 (8ms) or at 10,000 actions per second.

Each decision is written to a SHA-256 hash-chained audit log. That chain is tamper-evident: you can verify that no entry was deleted or altered after the fact. For compliance purposes, that is the difference between an audit log and a record.

What ships in the open-source package

The gateway ships six content inspection modules. These handle the categories of content and behavior that policy typically needs to check: data classification signals, prompt injection patterns, sensitive credential or secret exposure, and similar. The inspection runs before execution, not after.

The package also includes adapters for major hosted model APIs and full MCP support. If your agent stack uses MCP for tool calls, the proxy inserts into that path without an SDK migration. Vaikora supports 12 LLM providers in this way.

Self-hosting means your telemetry stays inside your perimeter. The agent requests, the policy decisions, and the audit chain do not leave your infrastructure unless you choose to send them somewhere.

Where the commercial control plane adds to it

The open-source proxy enforces policy locally. The Vaikora Control Plane, hosted by Data443, adds centralized policy management, compliance presets for HIPAA, PCI-DSS, GDPR, SOC 2, and ISO 27001, and the native integrations with Microsoft Sentinel, Microsoft Defender for Cloud, SentinelOne, and CrowdStrike Falcon.

The split is worth understanding. You can deploy vaikora-guard-mcp today, write your own policy rules, and get enforcement and audit logging without any commercial subscription. The control plane is where you manage policy at scale across multiple agents, map to compliance frameworks, and push high-severity decisions into your SIEM or XDR as incidents or threat intelligence indicators.

For teams that already use SentinelOne, Vaikora can push AI agent threats in as threat intelligence indicators, severity-mapped with Critical at 100, High at 75, Medium at 50, and Low at 25. No custom pipeline required.

Why the enforcement model is what matters

Monitoring what an agent did is useful for investigation. It does not stop the action. The enforcement-before-execution model means a policy violation that would have written a record to an external system, exfiltrated a file, or called a privileged API simply does not happen. The require-approval path adds a human decision for actions that fall outside automated allow or block territory without halting the agent entirely.

The audit chain gives you the ordered, signed record of every decision. That record is what a compliance audit or an incident investigation actually needs: not logs that might have been modified, but a chain that proves they were not.

What to check next

If you want to evaluate the enforcement logic before committing to any deployment:

• Install the package: `pip install vaikora-guard-mcp`
• Review the source on PyPI and inspect the content inspection modules against your own policy requirements
• Test a representative set of agent tool calls through the proxy and check the resulting audit chain entries
• If your agents connect to governed data sources, map your data classification policies to the proxy's inspection modules before moving to production

The open-source package is the starting point. What you are validating is whether the enforcement model fits the access patterns your agents actually have.

More from the Vaikora blog