VaikoraVaikora

Vaikora › Blog › Product Updates

Why we MIT-licensed the Vaikora MCP proxy and what that means for how you deploy policy enforcement

Product Updates · Nov 19, 2026 · 4 min read

Shipping an open-source enforcement layer is a specific architectural decision, not a marketing move. When you self-host vaikora-guard-mcp, every policy check runs in your own environment, against your own rules, with no agent action leaving your perimeter before a decision is made.

Here is what that actually changes for teams trying to enforce policy on AI agents.

What vaikora-guard-mcp is

vaikora-guard-mcp is a Python package, published on PyPI under the MIT license. It runs as an MCP-compatible proxy that sits inline between an AI agent and the tools it calls. Before any tool call executes, the proxy evaluates the action against your configured policy and returns one of four decisions: allow, block, modify, or require approval.

The enforcement point is before execution. If a call is blocked, the downstream tool never receives it. The agent sees the decision and must handle it. Nothing runs speculatively and gets cleaned up afterward.

The package includes six content inspection modules and adapters for the major hosted model APIs. It supports 12 LLM providers with no SDK migration required on your side.

Why self-hosting the enforcement layer matters

An enforcement proxy that you do not control is a policy dependency you cannot fully audit. Regulated teams in particular need to answer: where does the policy decision happen, what data crosses that boundary, and who can inspect the decision log?

With vaikora-guard-mcp self-hosted, the answers are: in your infrastructure, nothing, and you.

Every decision is written to a SHA-256 hash-chained, tamper-evident audit log. The chain is ordered and signed, so you can prove to an auditor that a specific tool call was blocked at a specific time, and that the log has not been modified since. That property does not require you to trust a third-party logging service.

This matters most in HIPAA, PCI-DSS, and GDPR contexts, where the question is not just "did you have controls" but "can you prove what the controls decided and when."

A concrete deployment scenario

Say you are running a code-generation agent that has access to a file-write tool and a shell-execution tool. The agent is built on a hosted LLM. You want to block any tool call that attempts to write to paths outside a defined directory, and require human approval before any shell command runs.

You install vaikora-guard-mcp from PyPI, configure two policy rules (a path-scope block and an approval trigger for shell calls), and point your agent's tool calls at the proxy. No changes to your LLM client, no SDK migration.

When the agent attempts a write to `/etc/hosts`, the proxy evaluates the path against your policy, produces a block decision in under 45ms at P99, signs the decision into the audit chain, and returns a rejection to the agent. The file system is never touched. When the agent later constructs a shell command, the proxy holds the call and routes it to your approval queue.

The agent cannot proceed on that branch until a human approves or rejects. If your on-call engineer rejects it, the rejection is in the same audit chain.

Open source gate, commercial control plane

vaikora-guard-mcp handles enforcement at the proxy layer. The Vaikora Control Plane, hosted by Data443, is the commercial product. It provides centralized policy management across multiple agents and deployments, the approval workflow UI, SIEM integrations (Microsoft Sentinel, CrowdStrike Falcon, SentinelOne, AWS Security Hub), and compliance presets for SOC 2, HIPAA, GDPR, PCI-DSS, and ISO 27001.

Teams that want to start with the open-source proxy can do so without any commercial commitment. The proxy enforces policy locally. When you need centralized governance across more than one agent deployment, or when your compliance program requires the SIEM integrations, the control plane connects without replacing or modifying the proxy you already deployed.

What to check before you deploy

Before you configure vaikora-guard-mcp, work through these questions:

The answers drive your initial policy configuration. Start with the highest-risk tool bindings. A block rule on two or three tools is more useful than an incomplete policy across twenty.

The package is on PyPI as `vaikora-guard-mcp`. Review the source, run it locally against a test agent, and check what the audit log produces before you treat any enforcement layer as production-ready.

See Vaikora enforce policy on your AI

Open-core AI runtime control. Self-host the MIT gateway free, or run the hosted Control Plane.

Get a demo Self-host the gateway

More from the Vaikora blog