Vaikora › Blog › Product Updates
vaikora-guard-mcp on PyPI: what the package actually contains and how to wire it into an MCP tool chain
vaikora-guard-mcp is a Python package, published on PyPI under the MIT license, that sits inline between an MCP client and the tools it calls. Before any tool executes, the proxy evaluates the call against policy and returns allow, block, modify, or require-approval. The source is public, self-hostable, and carries no SDK migration requirement.
This article walks through what the package contains, how enforcement works at the call level, and what you need to wire it into an existing MCP tool chain.
What ships in the package
Install with `pip install vaikora-guard-mcp`. The package ships six content inspection modules that run against each tool call payload before the call is forwarded. Each module targets a specific risk class: secrets in parameters, PII in arguments, policy scope violations, prompt injection patterns embedded in tool inputs, and similar categories.
The proxy also ships adapters for major hosted model APIs. If your agent already calls one of twelve supported LLM providers, no code migration is required. The proxy wraps the existing call path.
Every decision is written to a SHA-256 hash-chained audit log. Each entry is signed into the chain in order, so the record is tamper-evident. This matters for compliance: you get an ordered, verifiable sequence of what was allowed and what was blocked, not just a flat log file.
How a tool call moves through the proxy
A concrete example. An agent receives a task that requires calling a file-write tool. The agent constructs the tool call and sends it to the proxy. The proxy extracts the tool name, the arguments, and any context from the request.
The six inspection modules evaluate the payload. If a module flags a secret in the arguments (an API key passed as a plain string, for example), the proxy returns `block` to the agent before the file-write tool ever runs. The agent receives the block decision and the reason code. The file is never written.
If the call is clean, the proxy forwards it to the tool, receives the response, and returns it to the agent. The round-trip adds a P50 latency of 8ms and a P99 of 45ms. At scale, the proxy handles 10,000+ actions per second.
The enforcement decision is deterministic. Identical input produces an identical output every time. There is no probabilistic component in the enforcement path.
Compliance presets
The proxy ships with presets for SOC 2, HIPAA, GDPR, PCI DSS, and ISO 27001. A preset is a named policy bundle: a set of inspection rules, block conditions, and audit requirements that match the control surface each framework cares about.
For HIPAA, the preset flags tool calls that would pass PHI to an external endpoint not covered by the agent's configured data handling scope. For PCI DSS, it flags cardholder data patterns in tool arguments. You can extend any preset or write policy from scratch. The presets are starting points, not ceilings.
Connecting to the enterprise control plane
The open-source proxy is self-contained. Policy lives in a local configuration file and enforcement runs entirely within your infrastructure.
The optional Vaikora Control Plane, hosted by Data443, adds centralized policy management, cross-agent audit aggregation, and native integrations with Microsoft Sentinel, Microsoft Defender, SentinelOne, CrowdStrike Falcon, and AWS Security Hub. High-severity enforcement events push into those platforms automatically, with severity mapped at Critical=100, High=75, Medium=50, Low=25 for SentinelOne indicators.
The control plane is a commercial product. The proxy is MIT-licensed and free to use without it.
What to check before you deploy
Before wiring the proxy into a tool chain, review two things.
First, map the tool calls your agent already makes. The inspection modules need to cover the risk surface you actually have, not a generic default. If your agent calls internal APIs that handle regulated data, check whether the HIPAA or PCI preset covers the patterns those calls produce, or whether you need a custom rule.
Second, test the audit log against your compliance team's requirements. The SHA-256 chain gives you tamper evidence, but your team may need specific fields (agent ID, session context, tool name, decision reason) present in every entry. Confirm the default log schema matches before you treat it as a compliance artifact.
The package repository and configuration reference are on PyPI at `vaikora-guard-mcp`. Start there, run the proxy locally against a test tool, and inspect the audit log before connecting to production infrastructure.
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
Vaikora