Vaikora › Blog › Product Updates

Vaikora Guard MCP is now open source: what that means for policy enforcement on every agent tool call

Product Updates · Nov 6, 2026 · 3 min read

Every tool call an AI agent makes is a decision point. Without a gate, the agent decides. With Vaikora Guard MCP, a deterministic policy check runs before the tool call executes, and the result is always the same for the same input: allow, block, modify, or require approval.

That gate is now open source. `vaikora-guard-mcp` is published on PyPI under the MIT licence, written in Python, and self-hostable. The source is available for audit, modification, and integration without a commercial agreement.

What the proxy actually does

Vaikora Guard MCP sits inline between an MCP-compatible AI agent and the tools it calls. Before any tool call reaches execution, the proxy evaluates it against your policy configuration and returns one of four decisions: allow, block, modify, or require-approval.

The decision is deterministic. Identical input produces identical output. That is not a minor implementation detail. For compliance and audit purposes, it means you can explain every enforcement action by pointing to a policy rule, not an inference or a probabilistic score.

Every decision is written to a SHA-256 hash-chained, tamper-evident audit log. The chain gives you an ordered, verifiable record of every tool call your agent attempted and what the enforcement layer decided.

Why open source changes the trust equation

A policy gate sitting inline between your agent and your tools handles sensitive data at every evaluation. Requests that touch customer records, credentials, file contents, or external API calls all pass through it.

Publishing the source under MIT means your security team can read exactly what happens to that data at the inspection layer. You are not trusting a vendor's documentation. You are reading the code.

It also means you can self-host without a dependency on Data443's infrastructure. The commercial product is the Vaikora Control Plane, which handles policy management, the enterprise audit interface, and integrations with Microsoft Sentinel, CrowdStrike Falcon, SentinelOne, and AWS Security Hub. The gateway itself runs in your environment.

What ships with it

The open-source proxy includes six content inspection modules and adapters for the major hosted model APIs. It supports 12 LLM providers with no SDK migration required.

Compliance presets ship for SOC 2, HIPAA, GDPR, PCI DSS, and ISO 27001. For regulated environments, that means the policy structure for a specific framework is already defined. You configure the values for your environment; you do not build the rule set from scratch.

Enforcement latency at P50 is 8ms. At P99, it is 45ms. The gateway handles 10,000-plus actions per second. For teams concerned about adding a synchronous enforcement hop to agent workflows, those numbers define the actual overhead.

A concrete setup path

Your agent is running MCP tool calls today. You want a policy check before each one.

1. Install from PyPI: `pip install vaikora-guard-mcp`
2. Set your policy configuration file (or use a compliance preset)
3. Point your agent at the proxy endpoint instead of calling tools directly
4. The proxy evaluates each call, writes the decision to the audit log, and passes or blocks accordingly

If you are on AWS, Vaikora is available on AWS Marketplace as a SaaS subscription with private offer and EDP draw-down support. If your team runs Microsoft Sentinel, the Vaikora-AzureSecurityCenter solution ships in the Azure Sentinel content hub and promotes high-severity AI agent risk into Azure Security Center and Microsoft Defender for Cloud as native incidents.

What to check next

If you are evaluating this for a production deployment, the first question to answer is what your blocking policy looks like. The proxy can only enforce what you define.

Start by identifying which tool categories your agents call, what data they touch in those calls, and which calls should require human approval rather than automated allow or block. That mapping is your policy scope. Once you have it, the compliance presets give you a starting structure to modify rather than a blank configuration file.

The source is on PyPI. Read the inspection module code before you deploy it. That is exactly what open source is for.

More from the Vaikora blog