VaikoraVaikora

Vaikora › Blog › Product Updates

Vaikora Guard MCP is on PyPI: what it does to every tool call and why the license matters

Product Updates · Nov 20, 2026 · 4 min read

AI agents fail in a specific way. They call tools, read files, hit APIs, and write data, and nothing in the standard MCP stack stops a bad action before it executes. Vaikora Guard MCP is an open-source proxy that sits inline on that call path and enforces a policy decision before any tool call reaches its target.

What the proxy actually is

vaikora-guard-mcp is published on PyPI under the MIT license. It is written in Python and self-hosted. That means you can read every line, run it in your own environment, and inspect exactly what happens to a request before it moves forward.

The proxy sits between an AI agent and the LLM or tools it calls. Every action passes through the proxy. The proxy applies a policy check and returns one of four decisions: allow, block, modify, or require-approval. The action does not execute until one of those decisions is made.

That ordering matters. Detection after execution means the damage is already done. This is enforcement before execution.

What happens on each tool call

When an agent sends a tool call, the proxy:

Deterministic means identical input always produces an identical decision. There is no probabilistic scoring that lets the same prompt slip through on a second attempt.

Enforcement latency is P50 8ms and P99 45ms. At 10,000+ actions per second throughput, the proxy does not become the bottleneck in a production agent pipeline.

Every decision is written to a SHA-256 hash-chained, tamper-evident audit log. Each entry is signed into the chain, so you have an ordered, tamper-evident record of every allow, block, modify, and require-approval decision made across every agent session. That record is what compliance teams and investigators actually need.

Why the MIT license is a meaningful choice

Security teams are right to be skeptical of inline proxies that sit between their agents and their data. The MIT license means you can audit the enforcement code itself, not just the vendor's documentation about it. If your security team wants to verify that the proxy is not logging content it should not, they can read the source.

Self-hosted deployment means the proxy runs inside your own infrastructure. No request data leaves your environment to reach a third-party inspection service. For teams operating under HIPAA, PCI-DSS, GDPR, or similar requirements, that residency control is a practical requirement, not a preference.

Vaikora ships compliance presets for SOC 2, HIPAA, GDPR, PCI DSS, and ISO 27001. Those presets give a starting policy configuration that maps to the control requirements of each framework, reducing the time between installation and an auditable enforcement posture.

A concrete scenario

An agent is configured to read from a customer database and summarize records. Without a policy gate, the agent can read any record the underlying service account can access, including fields that contain payment card numbers or health identifiers.

With Vaikora Guard MCP inline, you write a policy rule that blocks any tool call response containing PAN or PHI patterns before the content reaches the agent context. The block is logged with a timestamp, the agent session ID, the tool that was called, and the policy rule that triggered. The agent sees a blocked response. No data leaves the tool. The audit log shows what was attempted and what stopped it.

That log entry is signed into the hash chain, so if an auditor or investigator asks what happened during that session, the record cannot be altered after the fact.

The enterprise control plane

The open-source proxy handles enforcement at the agent level. The Vaikora Control Plane, hosted by Data443, adds centralized policy management, cross-agent audit visibility, SIEM integrations with Microsoft Sentinel, CrowdStrike Falcon, SentinelOne, and AWS Security Hub, and the ability to push high-severity AI agent threats into those platforms as native incidents.

Vaikora is available on AWS Marketplace as a SaaS subscription and in the Azure Sentinel content hub as the Vaikora-AzureSecurityCenter solution. Teams that want self-contained enforcement with no external dependency can run on the open-source proxy alone. Teams that need centralized policy governance across many agents add the control plane.

What to check next

Install vaikora-guard-mcp from PyPI, point it at a non-production agent, and run a tool call that your current setup would allow. Check the audit log entry. Verify the policy decision that was applied and confirm the hash chain is intact. That test tells you whether your existing agent setup would have caught the same call in production.

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