Vaikora › Blog › Product Updates

MIT-licensed, self-hosted, nine-line setup: what you get when you open-source a policy gate for MCP agents

Product Updates · Nov 5, 2026 · 4 min read

Publishing source code under an MIT license shifts something specific for security teams: you can read exactly what runs between your agent and your tools, modify it, audit it, and deploy it in an environment you control. That is what the open-source release of Vaikora Guard MCP makes possible for anyone running MCP-based AI agents.

What the proxy actually is

Vaikora Guard MCP is a Python package, published on PyPI as `vaikora-guard-mcp`, that sits inline between an MCP agent and the tool calls it wants to execute. Before any tool call fires, the proxy evaluates it against a configured policy and returns one of four decisions: allow, block, modify, or require-approval. The action does not proceed until the decision is made.

That ordering matters. Detection after execution means the action already happened. This proxy runs the check first.

The gateway ships six content inspection modules and adapters for major hosted model APIs. It supports the Model Context Protocol natively. You self-host it. The commercial Vaikora Control Plane, hosted by Data443, is a separate product if you want centralized policy management, audit dashboards, and SIEM integrations. The proxy itself costs nothing and carries no usage restrictions under MIT.

What you configure and how policy applies

Policies are deterministic: identical input always produces an identical decision. That is not a soft claim. It means you can write a policy, test it against a fixture, and know the production behavior without probabilistic uncertainty.

Compliance presets ship for SOC 2, HIPAA, GDPR, PCI DSS, and ISO 27001. A team working under HIPAA, for example, can load that preset and immediately get content inspection rules that block PHI from leaving the agent context through a tool call. The preset defines what counts as a violation. The proxy enforces it before execution.

A worked example: an agent is processing a customer support ticket and decides to call a `send_email` tool. The agent passes the ticket body as context. The proxy inspects the payload, finds what the HIPAA preset flags as protected health information, and blocks the call. The agent receives a policy-violation response instead of a successful tool execution. The human operator sees the blocked call in the audit log.

The audit log and why it matters for investigations

Every decision the proxy writes to an audit log is signed into a SHA-256 hash chain. Each record links cryptographically to the one before it. Tamper-evident means you can demonstrate to an auditor or investigator that the log has not been modified after the fact.

For a SOC analyst reviewing an incident, this produces an ordered, verifiable record of what the agent attempted, what the policy decided, and when. That is a different quality of evidence than a standard log file.

Enforcement latency at P50 is 8ms and P99 is 45ms. At 10,000+ actions per second throughput, the proxy is not a bottleneck in production agent workflows.

What open source changes for security teams

The practical difference is auditability before deployment. A security engineer can read the inspection logic, trace how a content module flags a specific pattern, and confirm the behavior before putting it in front of production data. That is not possible with a closed binary.

The MIT license also means you can fork it, patch it, and ship a modified version internally without asking permission. If your environment has a specific data pattern or policy requirement not covered by the presets, you add it.

The package installs with `pip install vaikora-guard-mcp`. Setup requires nine lines of Python to have a working policy gate on tool calls. The gateway also ships as an MCP-compatible proxy, so if you are already running MCP infrastructure, it drops in without an SDK migration.

For teams that want to extend into enterprise policy management or push AI agent threat signals into Microsoft Sentinel, SentinelOne, CrowdStrike Falcon, or AWS Security Hub, Vaikora's control plane handles that. But the enforcement layer itself is open, self-hostable, and readable.

What to check next

Start with the PyPI page for `vaikora-guard-mcp` and read the source before you install it. Confirm which content inspection modules apply to your use case. If you have existing MCP agents in development or production, map which tool calls touch sensitive data and configure a policy for those first. The compliance presets are a reasonable starting point. If none of them fit, the MIT license means you write your own.

More from the Vaikora blog