VaikoraVaikora

Vaikora › Blog › Product Updates

vaikora-guard-mcp on PyPI: what the open-source MCP policy proxy does and how to run it

Product Updates · Nov 15, 2026 · 3 min read

When an AI agent calls a tool through the Model Context Protocol, the default behavior is execution. The call goes in, the function runs, and whatever happens next is a consequence of a decision no human reviewed. vaikora-guard-mcp changes that order: it sits inline and applies a policy check before the function executes.

The package is MIT licensed, written in Python, and published on PyPI as `vaikora-guard-mcp`. You can self-host it without a commercial agreement.

What it actually does

vaikora-guard-mcp acts as a proxy between an MCP client (your AI agent) and the MCP server (the tools the agent can call). Every tool call passes through the proxy. Before the call reaches the tool, the proxy evaluates it against policy and returns one of four deterministic decisions: allow, block, modify, or require-approval.

Deterministic here means what it says. Identical input produces identical output every time. There is no probabilistic scoring that lets a borderline call through on a good day and blocks it on a bad one.

The proxy ships with six content inspection modules. These cover things like data classification checks, pattern matching on arguments, and sensitive string detection, so you can write policy against what an agent is actually trying to pass to a tool.

Every decision is written to a SHA-256 hash-chained audit log. The chain gives you an ordered, tamper-evident record: you can prove not just what decision was made, but that the record has not been altered after the fact.

How to run it

Installation is a single pip command:

```
pip install vaikora-guard-mcp
```

From there, you define policy in a configuration file and point your MCP client at the proxy endpoint instead of directly at the tool server. The proxy handles the MCP protocol translation, so your agent code does not change.

A concrete scenario: an agent has access to a file-read tool and a file-delete tool. Without a proxy, a prompt injection in a document the agent reads could cause it to call file-delete on a path the attacker specifies. With vaikora-guard-mcp in place, you write a policy that blocks file-delete calls where the path argument matches directories outside a defined scope, or requires human approval for any deletion at all. The block happens before the function runs. There is no deletion to roll back.

What the enterprise control plane adds

The open-source proxy enforces policy locally. If you need centralized policy management across multiple agents or environments, that is where the Vaikora Control Plane comes in. It is a hosted commercial product from Data443 that manages policy across deployments, provides a UI for approval workflows, and aggregates the audit log across agents.

The proxy integrates natively with Microsoft Sentinel, Microsoft Defender, SentinelOne, CrowdStrike Falcon, and AWS Security Hub. AI agent threats can be pushed into those platforms as threat intelligence indicators with severity mapping, so a blocked tool call can appear as an incident in your existing SOC workflow without custom engineering.

Vaikora also ships as a SaaS subscription on AWS Marketplace and as a solution in the Azure Sentinel content hub.

Who this is for

If you run AI agents in production and those agents have tool access, any tool that writes data, calls an API, executes code, or deletes files is a potential enforcement point. The open-source proxy gives you a way to put policy at that point without waiting for a commercial procurement cycle.

The proxy works with 12 LLM providers with no SDK migration. Latency at P50 is 8ms and at P99 is 45ms, so the enforcement layer does not meaningfully change the agent's response time from an end-user perspective.

What to check next

Start by auditing what tools your agents currently have access to. For each tool, ask what the worst-case argument looks like and whether anything in your current setup would stop that call. If the answer is "we would see it in logs afterward," that is the gap the proxy addresses.

The source is on PyPI. Read the configuration reference before writing your first policy rule, specifically the section on argument inspection, because that is where most of the enforcement logic lives.

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