Vaikora › Blog › Product Updates
vaikora-guard-mcp is MIT-licensed and on PyPI: what that means for teams building policy into agent pipelines
The gap between "we have an AI agent" and "we have policy enforcement on that agent" is usually a build problem. Teams either wire up their own inspection layer, accept that tool calls run unchecked, or wait for a commercial product they can evaluate. vaikora-guard-mcp removes that choice by putting a self-hostable, open-source proxy on PyPI under the MIT licence, today, with no commitment required.
What vaikora-guard-mcp actually does
The proxy sits inline between an MCP-compatible AI agent and the tools it calls. Before any tool call executes, the proxy evaluates it against your policy and returns one of four decisions: allow, block, modify, or require-approval. The action either proceeds, is rewritten, is held for a human, or is stopped. Nothing executes before that decision is made.
That sequence matters because the alternative is detection after execution. If an agent exfiltrates a file, writes to a production database, or calls an external API it should not touch, a post-execution alert tells you what happened. A pre-execution gate stops it.
The enforcement path is deterministic. The same input always produces the same decision. There is no confidence score to tune, no threshold to debate in a post-incident review.
Installing and running it
```
pip install vaikora-guard-mcp
```
That installs the proxy. It is written in Python and published as `vaikora-guard-mcp` on PyPI. Because it is MIT-licensed and self-hosted, your team can read the source, fork it, and run it inside your own infrastructure without sending data to a third party.
For teams that need the commercial control plane, the Vaikora Control Plane is hosted by Data443 and adds centralized policy management, the SHA-256 hash-chained audit log, and the SIEM integrations. But the enforcement proxy itself is open.
A concrete scenario
Say you are running an AI agent that has access to a file-read tool and an HTTP-post tool. The agent decides, based on a prompt, to read a file containing a secret and post its contents to an external endpoint.
Without the proxy, that sequence executes. With vaikora-guard-mcp sitting inline, the HTTP-post call is evaluated before execution. If your policy blocks outbound calls to unrecognized endpoints, or blocks calls that carry content matching a secret pattern, the proxy returns a block decision and the call never fires.
You can configure the policy to modify the call instead, stripping the sensitive content before the post goes out. Or you can route it to a require-approval queue, where a human reviews the action before it proceeds.
What the proxy inspects
The proxy ships six content inspection modules. These cover the content of tool call arguments, not just the tool name or endpoint. That matters because a tool call to a legitimate endpoint can still carry policy-violating content.
The proxy also supports the major hosted model APIs, so the agent-side integration is a configuration change rather than an SDK migration.
Audit and compliance from the start
Every decision the proxy makes is written to a SHA-256 hash-chained, tamper-evident audit log. That log gives you an ordered, verifiable record of what the agent tried to do and what the policy decided, which is the foundation of any compliance argument.
Compliance presets for SOC 2, HIPAA, GDPR, PCI-DSS, and ISO 27001 ship with the platform, so teams working under those frameworks have a starting policy configuration they can adapt rather than build from scratch.
Why open source changes the evaluation path
Security teams are rightly skeptical of black-box enforcement claims. With an MIT-licensed proxy, the inspection logic is auditable by your team before you commit to anything. You can verify that a block decision on a tool call actually corresponds to the policy rule you think it does. That auditability is not a feature description. It is something you can confirm by reading the source.
For AI platform teams, the open-source proxy also means you can embed it directly into internal tooling, CI pipelines, or staging environments without a procurement process for the initial evaluation.
What to check next
If your team is already running MCP-compatible agents, install `vaikora-guard-mcp` from PyPI and run it against a test pipeline. Check whether your current tool calls would pass or fail the default policy configuration. That gap, between what your agents are doing and what your policy says they should do, is worth knowing before it shows up in an incident review.
The source is on PyPI. Start there.
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