Running an AI agent in production without a policy gate on its tool calls is an operational risk, not a theoretical one. The agent can read files, write to APIs, and call external services, and by default nothing checks whether any of those actions should be allowed before they execute. Vaikora Guard MCP is now published on PyPI under the MIT license to make it straightforward to put a deterministic policy check at that exact point.
What the Release Actually Is
`vaikora-guard-mcp` is a self-hosted, MCP-compatible proxy written in Python. It sits inline between an AI agent and the tools or LLMs it calls. Before any tool call executes, the proxy evaluates it against your policy configuration and returns one of four decisions: allow, block, modify, or require-approval. The action does not proceed until a decision is made.
The package is MIT licensed. You can install it, inspect the source, modify it, and run it on your own infrastructure. The commercial product is the Vaikora Control Plane, hosted by Data443, which adds fleet-wide policy management, the tamper-evident audit log, SIEM integrations, and compliance presets. The open-source proxy is a real, working enforcement component, not a demo wrapper.
Install from PyPI:
“`
pip install vaikora-guard-mcp
“`
Why Inline Enforcement at the Tool Call Layer Matters
The design choice to intercept at the MCP protocol layer, before execution, is what separates this from logging or monitoring approaches. If you are only recording what an agent did, you are building a forensic record. That is useful after an incident. It does not stop the write, the exfiltration, or the unintended API call.
A concrete example: an agent is working through a customer support workflow and receives a prompt that redirects it toward pulling records outside the original scope. With no gate in place, the tool call goes out. With `vaikora-guard-mcp` in the path, the call hits the proxy first. The proxy checks the action against configured policy, matches a data-access guardrail, and blocks the call before the downstream system receives the request. The agent gets a policy-denied response. The action never executes.
That is the enforcement model: deterministic, pre-execution, with an identical input always producing an identical decision.
What Ships in the Proxy
The open-source package ships six content inspection modules. These cover the content and structure of tool calls, not just their names, so a policy rule can match on what an agent is trying to send, not just which tool it is calling.
The proxy includes adapters for major hosted model APIs and full MCP support, so it can sit in front of agents built on different frameworks without requiring a migration. It works across twelve LLM providers. No SDK changes are needed on the agent side.
Enforcement latency is P50 8ms and P99 45ms, so the proxy does not materially change the response time profile of a typical agentic workflow.
What the Control Plane Adds
For teams that need governance at scale, the Vaikora Control Plane connects to the open-source proxy and adds:
- A SHA-256 hash-chained, tamper-evident audit log. Every decision is signed into an ordered chain that cannot be modified without detection.
- Compliance presets for HIPAA, PCI-DSS, GDPR, SOC 2, and ISO 27001, so you are not building policy rules from scratch for regulated use cases.
- Native integrations with Microsoft Sentinel, Microsoft Defender, SentinelOne, CrowdStrike Falcon, and AWS Security Hub, so AI agent risk surfaces in the same workflows your security operations team already uses.
- Throughput of 10,000+ actions per second, with 99.9% detection accuracy in testing.
The Vaikora Control Plane is available on AWS Marketplace as a SaaS subscription and in the Azure Sentinel content hub.
What to Check Next
If you are running AI agents today and have not put a policy gate in the tool-call path, start there. Install `vaikora-guard-mcp` from PyPI, review the source to understand how the inspection modules work, and configure a basic allow/block policy against one workflow before expanding scope.
If you need audit records that satisfy a compliance requirement, or you need to push AI agent risk into an existing SIEM, look at what the Control Plane adds on top of the open-source proxy.
The vaikora-guard-mcp repository is the right starting point. Policy rules, integration configuration, and the enforcement model are documented there.