VaikoraVaikora

Vaikora › Blog › Product Updates

Every MCP tool call runs through a policy check first: how Vaikora Guard MCP is built and why it's open source

Product Updates · Nov 19, 2026 · 4 min read

The practical risk in an AI agent is not the model itself. It is the moment the agent decides to call a tool: read a file, query a database, send a request to an external API. That is where a policy check either exists or does not.

Vaikora Guard MCP puts that check inline, before execution, on every tool call. The package is MIT-licensed, written in Python, and published on PyPI as `vaikora-guard-mcp`. You can read the source, audit the enforcement logic, and self-host the proxy without a commercial agreement.

What "inline before execution" means in practice

Most agent observability tools record what an agent did. Vaikora Guard MCP runs between the agent and the tool, intercepts the call, evaluates it against policy, and returns an allow, block, modify, or require-approval decision before the tool executes.

The enforcement path looks like this:

The proxy adds a P50 latency of 8ms and a P99 of 45ms. At that speed, enforcement is not a meaningful bottleneck for most agentic workflows.

Why deterministic policy matters for auditability

Non-deterministic controls create audit problems. If two identical inputs can produce different decisions, you cannot reconstruct what happened or prove a control was operating consistently during an investigation.

Vaikora Guard MCP enforces policy deterministically. Every decision is written to a SHA-256 hash-chained, tamper-evident audit log. The chain gives you an ordered, verifiable record: what the agent requested, what policy matched, what decision was made, and when. That record is what compliance teams and investigators actually need.

This is especially relevant if you operate under HIPAA, PCI-DSS, GDPR, SOC 2, or ISO 27001. Vaikora ships compliance presets for each of those frameworks, so you are not writing policy from scratch for every regulated workflow.

A concrete scenario: an agent with file system and API access

Say you have an agent that answers internal questions by reading from a shared file store and calling an external enrichment API. Without a policy layer, the agent can read any file it has filesystem access to and send any content to the external API. The access control question is entirely delegated to the underlying permissions.

With Vaikora Guard MCP in the path:

This is what it means to add a runtime gate on top of existing IAM and RBAC controls rather than replacing them. The underlying permissions still apply. Vaikora adds an additional enforcement layer above them.

Why MIT-licensed and self-hostable

The enforcement proxy, `vaikora-guard-mcp`, is the open-source component. The Vaikora Control Plane, which manages policy centrally across agents, teams, and environments, is the commercial product hosted by Data443.

Releasing the proxy under MIT means security teams can inspect exactly what runs inline in their agent traffic. For teams with strict requirements about third-party code in sensitive environments, self-hosting the proxy is a practical option without needing to trust a black box.

It also means the proxy can support 12 LLM providers with no SDK migration. The gateway adapts to the provider, not the other way around.

What to check next

If you want to evaluate the enforcement behavior before connecting a control plane, install `vaikora-guard-mcp` from PyPI and read the source. The six inspection modules are visible, the policy evaluation path is traceable, and the audit log format is documented.

The specific thing to verify for your environment is whether the tool calls your agents make are the calls you think they are making. Instrument the proxy against a representative workload and review the audit log. What the agent actually requests and what your policy would have allowed is often not the same list.

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

More from the Vaikora blog