Vaikora › Blog › Product Updates

How Vaikora Guard MCP enforces policy on every agent tool call before the call executes

Product Updates · Nov 10, 2026 · 4 min read

The gap in most AI agent deployments is not detection. It is timing. Agents call tools, write records, and trigger external services before any review can happen. By the time a log entry appears, the action is done.

Vaikora Guard MCP is the open-source component that closes that gap. It sits inline between an agent and the tools it calls, intercepts each request, evaluates it against policy, and either allows, blocks, modifies, or routes it for human approval. That decision happens before execution.

What it is and where to get it

Vaikora Guard MCP is published on PyPI as `vaikora-guard-mcp`. It is written in Python, licensed under MIT, and self-hosted. That means your team can read the code, run it in your own environment, and inspect exactly what policy is applied to each call.

It works as an MCP (Model Context Protocol) proxy. Agents connect to it the same way they would connect to any MCP server. No SDK migration, no instrumentation changes in your agent code. The proxy intercepts the tool call at the protocol layer.

What happens to each tool call

When an agent sends a tool call through the proxy, Vaikora Guard MCP runs it through content inspection before forwarding it. The inspection evaluates the call against whatever policy rules are configured: data type, action category, destination, sensitivity classification, or any custom condition your team defines.

The result is deterministic. Identical input always produces an identical policy decision. That property matters for auditability: you can trace any blocked or modified action back to the exact rule that triggered it.

Every decision is written to a SHA-256 hash-chained audit log. Each entry is signed into the chain, so the record cannot be altered after the fact. For compliance purposes, this gives you an ordered, tamper-evident account of every action your agents attempted.

A concrete example

Say an agent is authorized to read from a customer database and summarize records. It calls a tool that happens to include a `DELETE` operation in the same schema. Without a pre-execution check, that call might succeed depending on how the downstream role is configured.

With Vaikora Guard MCP in the path, the proxy inspects the tool call before it reaches the database. If your policy blocks write or delete actions for that agent context, the call is blocked at the proxy. The agent receives a denial, the action never reaches the database, and the decision appears in the audit log with a timestamp and the matched rule.

This is different from catching the event in a SIEM after the fact. The row is never deleted.

How it connects to the control plane

The open-source proxy enforces policy locally. For teams that need centralized policy management across multiple agents or environments, Data443's Vaikora Control Plane is the commercial product that hosts above the proxy. The control plane distributes policy to each Guard MCP instance, aggregates audit logs, and provides the management interface.

The proxy and the control plane are separate. You can run the proxy standalone with local policy files, or connect it to the control plane for fleet-wide governance.

Throughput and latency

The enforcement path is designed to stay out of the way of agent performance. Vaikora handles over 10,000 actions per second, with P50 latency at 8ms and P99 at 45ms. For most agent workflows, that overhead is not perceptible.

Compliance presets

The proxy ships with presets for SOC 2, HIPAA, GDPR, PCI DSS, and ISO 27001. These are starting configurations, not locked rules. Each preset defines a baseline of what gets blocked or flagged for that regulatory context, and your team can adjust or extend the rules from there.

What to do next

If you want to inspect the code before deploying anything, `pip install vaikora-guard-mcp` pulls the package from PyPI. The source is there to read.

If you have agents already running against MCP servers, check whether any tool call in your current setup reaches a write, delete, or external send operation without an inline policy check in the path. That is the first place to route traffic through the proxy.

If you need the control plane for multi-agent environments, it is available on AWS Marketplace and in the Azure Sentinel content hub.

More from the Vaikora blog