Vaikora › Blog › Product Updates

Vaikora Guard MCP is now open source: here is what it does to every agent tool call before execution

Product Updates · Nov 11, 2026 · 4 min read

When an AI agent calls a tool, the call either gets checked before it executes or it does not. Most current agent frameworks do not check. Vaikora Guard MCP is a self-hostable, MIT-licensed proxy published on PyPI as `vaikora-guard-mcp` that sits inline between an agent and its tools and enforces a policy decision before each call goes through.

This post explains what the proxy actually does, how the enforcement works, and what you get from open-sourcing it.

What the proxy intercepts

Vaikora Guard MCP implements the Model Context Protocol (MCP) as a proxy layer. The agent sends a tool call to the proxy. The proxy inspects the call against configured policy, produces one of four decisions (allow, block, modify, or require-approval), and either forwards the call, returns a block, rewrites the payload, or holds it for a human. The underlying tool never sees a call that the policy rejected.

A concrete example: an agent is working through a customer support workflow and calls a `read_file` tool with the path `/etc/passwd`. The proxy catches that path pattern against a policy rule, blocks the call, and returns a structured error to the agent. The file system is never touched. The agent receives the block response and can be instructed to handle it gracefully in its system prompt.

The enforcement is deterministic. Identical input always produces an identical decision. This matters for auditability: you can replay a log entry and confirm what the proxy would have decided at that moment.

How decisions are recorded

Every decision the proxy makes is written to a SHA-256 hash-chained, tamper-evident audit log. Each entry references the previous entry's hash, so any modification to a past record breaks the chain and is detectable. For compliance programs under SOC 2, HIPAA, GDPR, PCI-DSS, or ISO 27001, this gives you an ordered, verifiable record of every action an agent attempted and what happened to it.

This is different from application logging. A standard log can be truncated, overwritten, or silently dropped. The hash chain makes tampering visible. When an auditor or incident responder asks what an agent did between 14:00 and 14:30, the answer comes from a record that cannot be quietly edited after the fact.

Enforcement latency at P50 is 8ms, P99 is 45ms. For most agent workflows, this is below perceptible delay.

What open source changes operationally

Publishing the proxy under MIT means security teams can read, audit, and modify the enforcement code without asking for access. If your security policy requires inspecting every component that sits inline with production traffic, the full source is available.

The package installs from PyPI:

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

Self-hosting means the proxy runs in your own environment. Tool call content never transits a third-party service to reach the enforcement layer. For regulated workloads where data residency or egress controls matter, this is not optional.

The open-source proxy connects to the Vaikora Control Plane, which is the commercial product hosted by Data443. The control plane is where you manage policy centrally, review approval queues, and access the audit trail across multiple agents and environments. The proxy is the enforcement point; the control plane is the management layer.

Connecting to your existing security stack

The proxy's decisions feed into the broader detection and response workflow. Vaikora integrates natively with Microsoft Sentinel, Microsoft Defender, SentinelOne, CrowdStrike Falcon, and AWS Security Hub. High-severity agent threats are pushed to SentinelOne as threat intelligence indicators with severity mapping (Critical=100, High=75, Medium=50, Low=25) and promoted into Azure Security Center and Microsoft Defender for Cloud as native incidents with no custom engineering required.

This means a blocked tool call does not stay siloed in an agent log. A policy violation at P99 latency can appear in your SIEM before the next agent step executes.

Compliance presets

The proxy ships with presets for HIPAA, PCI-DSS, GDPR, SOC 2, and ISO 27001. A preset is a starting-point policy configuration that covers the most common rules for that framework: for example, blocking tool calls that would read or write fields matching PCI-DSS cardholder data patterns. Teams can extend or override presets with their own rules.

These presets reduce the time from "we need AI agent controls for our compliance program" to "we have a running enforcement configuration" to a single afternoon rather than a multi-week policy design project.

What to check next

If you are running agents against MCP-compatible tools today, install `vaikora-guard-mcp` from PyPI and run it in front of a non-production tool. Review the audit log after a short session and check whether the tool calls your agent is making match what you expected it to do. The gap between expected and actual behavior is usually where the first useful policy rule comes from.

For production use, review the hash-chained audit output format and confirm it integrates with your existing log pipeline before you need it for an incident.

More from the Vaikora blog