The Model Context Protocol gives AI agents a standard way to call tools. It does not give you any control over which calls are allowed, what data can pass through, or what happens when a call violates a policy you care about. That gap is what Vaikora Guard MCP is built to close.
The project is published on PyPI as `vaikora-guard-mcp`, written in Python, and licensed MIT. You can read the source, run it locally, modify it, and deploy it in your own environment without any dependency on a hosted service.
What the Proxy Actually Does
Vaikora Guard MCP sits inline between an AI agent and the tools it calls via MCP. Before a tool call executes, the proxy evaluates it against a policy definition. The result is one of four decisions: allow, block, modify, or require-approval. The call either proceeds, is stopped, is rewritten, or is held for a human to review.
This is deterministic enforcement. Identical input produces an identical decision every time. There is no probabilistic scoring or model-based judgment involved in the policy gate itself.
Every decision is written to a SHA-256 hash-chained audit log. The chain is tamper-evident: you can verify that the record has not been altered after the fact, which matters when you need to demonstrate to an auditor that a specific tool call was blocked or approved at a specific time.
Why Inline Matters Here
A common pattern for AI agent governance is logging tool calls and reviewing them afterward. That approach gives you visibility. It does not give you control, because the action has already run by the time you see it.
Inline enforcement changes the order of operations. The policy check happens before execution. If an agent tries to call a tool that would write a record containing PII to an external endpoint, the block fires before the write happens. The data never leaves.
Consider a concrete scenario: an AI coding agent is authorized to query a read-only database view. A prompt injection in a retrieved document causes the agent to attempt a DELETE call against a production table. Without an inline gate, that call goes to the tool and executes. With Vaikora Guard MCP in the path, the call is evaluated against policy first. A rule that permits SELECT and blocks anything else stops the DELETE before it reaches the database.
What the Open-Source Package Contains
The `vaikora-guard-mcp` package ships with:
- Six content inspection modules covering categories such as PII detection, secret scanning, and prompt injection patterns
- Adapters for the major hosted model APIs, so the proxy works without changing SDK code in the agent
- MCP support, meaning it integrates with any agent framework that speaks the Model Context Protocol
- Policy definition in a format you can read and version-control alongside your agent code
Compliance presets for SOC 2, HIPAA, GDPR, PCI-DSS, and ISO 27001 are included. Each preset is a starting-point policy that maps to the control requirements of that framework. You can extend or restrict it.
The gateway enforces policy with a P50 latency of 8ms and P99 of 45ms. For most tool call workflows, the enforcement overhead is not perceptible.
The Control Plane Layer
The open-source proxy handles runtime enforcement. If you need centralized policy management across multiple agents, audit log aggregation, or the SIEM integrations (Microsoft Sentinel, CrowdStrike Falcon, SentinelOne, AWS Security Hub), those come from the Vaikora Control Plane, which is the commercial product hosted by Data443.
The split is intentional. The enforcement logic runs in your environment, under your control, with no required call home. The Control Plane adds the management layer on top.
Vaikora is available on AWS Marketplace as a SaaS subscription and ships in the Azure Sentinel content hub as the Vaikora-AzureSecurityCenter solution, for teams that want the full stack with native cloud integrations.
What to Check Next
If you are running agents in production today, the first thing to verify is whether any tool call path currently operates without a policy gate. Look specifically at:
- Calls that write to external systems or databases
- Calls that read data with PII or secrets in scope
- Any agent that accepts retrieved content from outside your trust boundary, since that is the injection vector
The source for `vaikora-guard-mcp` is on PyPI. Install it, read the policy schema, and map your existing tool permissions against what the default inspection modules would catch. That gap analysis is the starting point.