Vaikora › Blog › Product Updates

pip install vaikora-guard-mcp: what runs between your agent and its tools, and why it matters

Product Updates · Nov 8, 2026 · 4 min read

Every tool call an AI agent makes is a decision point. Without a policy gate at that point, the agent's judgment is the only control in place. That is not a security posture.

Vaikora Guard MCP is the open-source component that sits inline between an MCP-compatible AI agent and the tools it calls. It is MIT-licensed, written in Python, and published on PyPI as `vaikora-guard-mcp`. You can install it, read all of it, and run it in your own environment without touching the enterprise control plane.

What "inline" actually means here

When an agent issues a tool call, the request goes to the Vaikora Guard MCP proxy first. The proxy inspects the call against your configured policy and returns one of four decisions: allow, block, modify, or require-approval. Only after an allow or modify decision does the call reach the actual tool.

This is enforcement before execution, not detection after the fact. If a policy says an agent cannot exfiltrate PII to an external endpoint, that call never completes. There is no race condition between detection and action.

The proxy is MCP-compatible, which means it works with any MCP-aware agent framework without SDK migration. You point your agent at the proxy instead of the tool directly.

What the proxy inspects

The gateway ships six content inspection modules. These cover things like sensitive data patterns in payloads, destination controls, action type restrictions, and prompt injection signals. Each module runs against the tool call before the decision is finalized.

Identical input always produces an identical decision. The enforcement logic is deterministic, not probabilistic. That matters for auditability: when a compliance team asks why a specific call was blocked at 14:32 on a given date, the answer is a policy rule and a payload hash, not a confidence score.

Every decision is written to a SHA-256 hash-chained, tamper-evident audit log. Each entry is signed into an ordered chain, so the record cannot be altered after the fact without breaking the chain. That audit trail is what makes the system usable for regulatory evidence, not just operational telemetry.

A concrete scenario

Say an agent is authorized to query a customer database and summarize records for a support workflow. Without a policy gate, nothing stops the agent from being prompted (by a user or by an injected instruction in a document it reads) to dump the full table to an external webhook.

With Vaikora Guard MCP in place, that tool call hits the proxy. The content inspection modules flag the external destination and the volume of records in the payload. The policy is "block any outbound tool call containing more than N customer records to a non-approved destination." The call is blocked. The decision, the payload hash, and the policy rule that triggered the block are written to the audit log. The agent receives a structured rejection and stops.

The support workflow continues for legitimate calls. The control applies only where the policy says it should.

Running it

Install from PyPI:

```
pip install vaikora-guard-mcp
```

Configure your policy rules in the proxy config (YAML-based, documented in the repo). Point your MCP agent at the proxy address instead of the tool's address directly. The proxy handles the MCP handshake transparently.

For teams that want centralized policy management across multiple agents, the Vaikora Control Plane is the commercial layer. It is hosted by Data443 and adds policy versioning, approval workflows, a dashboard over the audit log, and the SIEM integrations (Microsoft Sentinel, CrowdStrike Falcon, SentinelOne, AWS Security Hub). The open-source proxy is designed to connect to that control plane when you need it, but it runs standalone without it.

The gateway is also available on AWS Marketplace as a SaaS subscription and in the Azure Sentinel content hub as the Vaikora-AzureSecurityCenter solution, which matters for teams whose procurement and deployment already go through those channels.

What to check next

If you are evaluating agent security controls right now, the most useful thing you can do is read the inspection module source before you deploy anything. The repo is public. Look at what each module checks, map it to the tool calls your agents actually make, and decide where you need to add policy rules that are not in the defaults.

If you are already running MCP-compatible agents in production without an inline policy gate, that is the gap to close first.

More from the Vaikora blog