Vaikora › Blog › Product Updates

Vaikora Guard MCP is open source: how the MIT-licensed proxy puts policy before every agent tool call

Product Updates · Nov 12, 2026 · 4 min read

The problem with most AI agent guardrails is timing. The check happens after the tool call completes, or after output is returned, or inside a wrapper that only catches what the developer thought to handle. By then, the action has already executed.

Vaikora Guard MCP is a self-hosted, MIT-licensed, Python proxy published on PyPI as `vaikora-guard-mcp`. It sits inline between an AI agent and every tool it calls. The policy decision runs before the call executes. That ordering is the point.

What open source actually means here

The MIT license means you can read the full enforcement logic, modify it, audit it, and self-host it without a commercial agreement. The source is on GitHub. The package installs from PyPI with a standard `pip install vaikora-guard-mcp`. There is no SDK migration required, and because the proxy is MCP-compatible, it works with the agent frameworks that already speak that protocol.

The commercial product is the Vaikora Control Plane, hosted by Data443. The gateway itself, including the six content inspection modules and the adapters for major hosted model APIs, is what you get for free under MIT.

This structure matters for security teams with strict supply-chain requirements. You can inspect every line of enforcement logic before it touches production traffic.

What happens at the proxy layer

When an agent issues a tool call, Vaikora Guard MCP intercepts the request before it reaches the LLM or external tool. It runs the request through deterministic policy: identical input always produces identical output. The result is one of four decisions: allow, block, modify, or require approval.

The proxy ships six content inspection modules. Those handle things like prompt injection detection, sensitive data patterns, and instruction boundary violations. Each decision is written into a SHA-256 hash-chained audit log, so the record is tamper-evident and ordered.

Latency at P50 is 8ms and at P99 is 45ms. At 10,000+ actions per second throughput, the proxy does not become the bottleneck in a production agent pipeline.

A concrete scenario

Say an agent is processing a support ticket. The ticket contains what looks like an internal system path and a request to read a file. Without a runtime gate, the agent calls the file-read tool and the action executes. With Vaikora Guard MCP inline, the content inspection module flags the path pattern before the tool call fires. Policy says block. The call never reaches the tool. The decision lands in the audit log with a timestamp and hash.

The same logic applies to database writes, external API calls, credential lookups, or any other tool the agent has access to. The proxy enforces across all of them, not just the ones a developer explicitly wrapped.

Compliance presets and SIEM integration

The open-source proxy connects to the Vaikora Control Plane for policy management and reporting. The control plane ships presets for HIPAA, PCI-DSS, GDPR, SOC 2, and ISO 27001. For teams running Sentinel or CrowdStrike, Vaikora integrates natively with Microsoft Sentinel, Microsoft Defender, SentinelOne, CrowdStrike Falcon, and AWS Security Hub.

That means AI agent threat signals produced by the proxy can flow directly into the SIEM and EDR platforms a SOC already operates. No custom engineering is required to get Vaikora incidents into Azure Security Center or Microsoft Defender for Cloud. High-severity decisions push into SentinelOne as Threat Intelligence indicators with severity scores mapped at Critical=100, High=75, Medium=50, and Low=25.

What Vaikora adds on top of existing IAM

One thing worth stating plainly: Vaikora Guard MCP does not replace Azure RBAC, AWS IAM, or Snowflake roles. It adds a runtime gate on top of them. Those systems control what an identity is permitted to access. Vaikora controls what an AI agent is permitted to do at the action level, in real time, before execution. Both layers are doing different work.

Where to start

Install `vaikora-guard-mcp` from PyPI and read the source before deploying it anywhere. The MIT license means nothing is hidden. Check the content inspection module configuration against the tool permissions your agents currently hold. That gap, between what an agent can call and what it should be allowed to call under policy, is where enforcement needs to live.

If your agent framework speaks MCP, the proxy drops in without a rewrite. If you need the control plane for policy management, audit reporting, or compliance preset configuration, that is available as a SaaS subscription on AWS Marketplace with support for private offers and EDP draw-down.

More from the Vaikora blog