Every tool call an AI agent makes is a decision point. Without a gate sitting inline at that point, the agent executes first and you find out what happened after. Vaikora Guard MCP is the open-source component that puts a deterministic policy check at that exact moment, before the tool call completes, and it is now available on PyPI under the MIT license.
What the Open-Source Release Actually Includes
The package, `vaikora-guard-mcp`, is an MCP-compatible proxy written in Python. It sits between an AI agent and whatever LLM or tool the agent is calling, and it applies a policy decision before each action executes. The decision is one of four: allow, block, modify, or require approval. That decision is applied before execution, not logged after.
The MIT license means you can run it in your own environment, audit the code, and integrate it into existing pipelines without a commercial agreement. It is self-hostable by design. The open-source release covers the gateway itself. The enterprise control plane, hosted by Data443, is the commercial layer where you manage policy at scale, review audit chains, and push signals into your SIEM and EDR tooling.
You can install the package directly:
“`
pip install vaikora-guard-mcp
“`
From there, the proxy intercepts tool calls and runs them through the configured policy rules before the agent proceeds.
How the Policy Check Works at Runtime
When an agent issues a tool call, the request hits the Vaikora Guard proxy. The proxy runs the request through its content inspection modules, checks it against the active policy set, and returns a decision. The agent receives either approval to proceed, a block with a reason, a modified version of the request, or a hold pending human approval.
Identical input always produces an identical decision. There is no probabilistic scoring that lets a prompt slip through one run and get blocked the next. The policy is deterministic. This matters for auditability: if a decision is challenged in a compliance review or an incident investigation, the log entry reflects the exact rule that fired and the exact decision that was made.
A concrete example: an agent is working through a workflow that involves querying a customer database and then writing a summary to an external endpoint. The tool call to write to the external endpoint contains PII from the query result. The proxy inspects the outbound payload, matches it against a data exfiltration rule, and blocks the action. The agent receives a block response. No data leaves. The event is written to the audit log before the agent can try again.
Enforcement latency at P50 is 8ms and at P99 is 45ms. At 10,000 or more actions per second, the gate does not become the bottleneck in the pipeline.
What the Audit Log Covers
Every policy decision, allow or block, is written to a SHA-256 hash-chained, tamper-evident audit log. Each entry references the previous entry’s hash, so any modification to the log is detectable. This gives you an ordered, verifiable record of what every agent did, what policy applied, and what was permitted or stopped.
For compliance teams working against HIPAA, PCI-DSS, GDPR, SOC 2, or ISO 27001, Vaikora ships presets for each of these frameworks. The audit chain produced by the open-source gateway is the same format used by the enterprise control plane, so records produced in a self-hosted deployment carry the same evidentiary structure.
Connecting the Gateway to Your Existing Stack
The open-source gateway works with 12 LLM providers and requires no SDK migration to add to an existing agent deployment. The enterprise control plane extends that with native integrations into Microsoft Sentinel, Microsoft Defender, SentinelOne, CrowdStrike Falcon, and AWS Security Hub.
For teams running agents on AWS, Vaikora is available on AWS Marketplace as a SaaS subscription. For Azure environments, 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 without custom engineering.
Vaikora also adds a runtime gate on top of existing identity controls (Azure RBAC, AWS IAM, Snowflake roles) rather than replacing them. The enforcement layer is additive.
What to Check Before You Deploy
Before running the proxy in a production agent pipeline, review these:
- Policy scope: which tool call types are in scope and which actions should require human approval versus automatic block
- Content inspection modules: the gateway ships six inspection modules; confirm which ones apply to your data types and agent workflows
- Audit log retention: decide where the hash-chained log is stored and for how long, based on your compliance obligations
- Escalation path: for actions that return a require-approval decision, define the approval workflow before agents start queueing requests
The source code is on PyPI at `vaikora-guard-mcp`. Review the inspection module configuration first, then map your agent workflows to the four decision types before writing policy rules.