Vaikora › Blog › Product Updates
Vaikora Guard MCP is now open source: here is what it does before each agent tool call runs
The Model Context Protocol has become the standard wiring between AI agents and the tools they call. That means any gap in policy enforcement at the MCP layer is a gap that sits directly in front of file writes, database queries, API calls, and shell commands. Vaikora Guard MCP closes that gap by acting as a proxy between the agent and those tools, applying a policy decision before each call executes.
The package is MIT-licensed, written in Python, and published on PyPI as `vaikora-guard-mcp`. You can self-host it today without touching any of the AI providers or tools you already use.
What the proxy actually does
Vaikora Guard MCP sits inline on the MCP message path. When an agent constructs a tool call, the proxy intercepts it before it reaches the tool server. It runs the call through Vaikora's content inspection modules, evaluates it against your configured policy, and returns one of four decisions: allow, block, modify, or require-approval.
That decision is deterministic. Identical input produces identical output every time, which matters when you need to demonstrate consistent enforcement to an auditor. Every decision is also 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. For most agent workflows, that is not perceptible to the user and does not change the tool call pattern.
A concrete scenario
Say an agent is authorized to read from a customer database to populate a support ticket. The same agent also has access to a tool that sends outbound email. Without a policy gate, nothing stops the agent from calling the email tool with data it just retrieved from the database, because both tool calls are individually permitted.
With Vaikora Guard MCP in place, you write a policy that blocks any outbound communication tool call that follows a database read containing fields tagged as PII. The proxy sees the sequence, evaluates the second call against that policy, and blocks it before the email is sent. The block is logged with the full context of the tool call, the policy that fired, and a timestamp in the audit chain.
That kind of cross-call context is what separates runtime enforcement from static permission models. The agent's MCP credentials may allow both tools. The policy enforces what the combination of calls is allowed to do.
Configuration and compliance presets
The gateway ships six content inspection modules. You configure which modules run against which tool call types, and you set the policy logic per module. For teams that need to demonstrate compliance, Vaikora ships presets for SOC 2, HIPAA, GDPR, PCI DSS, and ISO 27001. A preset is a starting policy configuration, not a certification claim. You still need to validate that the policy correctly covers your specific data and tool surface.
The open-source proxy connects to the Vaikora Control Plane for policy management and audit log storage. The Control Plane is the commercial product, hosted by Data443. The proxy itself carries no licensing cost.
How it fits into your existing stack
Vaikora Guard MCP supports 12 LLM providers with no SDK migration. If your agents already call tools through MCP, adding the proxy is a configuration change, not an architectural one.
For teams already running security operations in SIEM or endpoint platforms, the proxy feeds into that existing telemetry. Vaikora integrates with Microsoft Sentinel, Microsoft Defender, SentinelOne, CrowdStrike Falcon, and AWS Security Hub. AI agent threats can surface as native incidents in those platforms without custom engineering.
The proxy does not replace existing IAM controls. If your agent operates under AWS IAM roles or Azure RBAC, those permissions still apply. The proxy adds a policy gate on top of them, enforcing at the application layer what IAM enforces at the credential layer.
What to check before you deploy
- Map every MCP tool your agents call. The proxy only enforces what it intercepts. Tools called outside the MCP path are outside its scope.
- Review which tool call combinations carry actual risk. The most useful policies govern sequences and combinations, not just individual calls.
- Decide whether you need the require-approval path for high-risk tools. That path pauses execution and waits for a human decision before the call proceeds.
- Pull the package from PyPI (`pip install vaikora-guard-mcp`), review the source, and run it against a staging environment before connecting a production tool surface.
The source is on GitHub. Start with the proxy configuration documentation to understand how inspection modules map to tool call types before you write your first policy.
See Vaikora enforce policy on your AI
Open-core AI runtime control. Self-host the MIT gateway free, or run the hosted Control Plane.
Get a demo Self-host the gateway
Vaikora