Vaikora › Blog › Product Updates
Vaikora Guard MCP is now open source: here is what that means for policy enforcement on every tool call
The MIT-licensed release of vaikora-guard-mcp puts a deterministic policy gate directly in the path of every MCP tool call, before execution, with no SDK migration required. It is available today on PyPI as `vaikora-guard-mcp` and self-hostable by any team running MCP-compatible agents.
This post covers what the proxy actually does, how the policy checks work, and what you get out of the box before connecting the commercial control plane.
What the proxy does, specifically
`vaikora-guard-mcp` sits inline between an AI agent and the tool or model it is about to call. When the agent issues a tool call, the proxy intercepts the request, evaluates it against configured policy, and returns one of four decisions: allow, block, modify, or require-approval. The underlying tool call only executes if the proxy permits it.
This is not a wrapper that logs after the fact. The enforcement happens before execution. If a policy rule says block any tool call that includes a string matching a Social Security Number pattern, the exfiltration attempt never reaches the downstream function.
The proxy ships with six content inspection modules. These cover categories like sensitive data patterns, prompt injection markers, and tool call structure. You configure which modules apply and what they trigger.
Why "before execution" is the design that matters
Consider a concrete scenario. An AI coding agent is given access to a file-read tool and a network-post tool. A prompt injection embedded in a file instructs the agent to read `/etc/passwd` and post its contents to an external endpoint. Without an inline gate, the agent executes both calls in sequence. The only record is whatever your LLM provider logs, if you have that enabled.
With `vaikora-guard-mcp` in path, the read call hits the proxy first. If the file path matches a blocked pattern, the action stops there. The network-post call never fires. Every decision, including the block, is written to a SHA-256 hash-chained audit log, so you have a tamper-evident record of what was attempted and what policy rejected it.
Identical input always produces an identical decision. That determinism matters for compliance investigations. You can replay the log and show exactly why a given action was blocked or allowed.
What ships in the open-source package
The MIT-licensed release includes:
- The MCP-compatible proxy core
- Six content inspection modules
- Adapters for the major hosted model APIs, covering 12 LLM providers
- Policy configuration in YAML
- The SHA-256 hash-chained audit log
- Self-hosting support with no external dependency on the Data443 control plane
Compliance presets for SOC 2, HIPAA, GDPR, PCI DSS and ISO 27001 ship with the package. Each preset maps to a specific set of content rules and block/allow defaults, so you are not writing policy from scratch for a regulated environment.
Enforcement latency at P50 is 8ms and P99 is 45ms. For most agent workflows, that is below the threshold where users or downstream systems notice it. The proxy handles 10,000 or more actions per second, so it does not become a bottleneck in high-volume pipelines.
What the enterprise control plane adds
The open-source proxy enforces policy locally, using whatever configuration you provide. The Vaikora Control Plane, hosted by Data443, adds centralized policy management across multiple agents, real-time visibility into decision telemetry, and native integration with Microsoft Sentinel, Microsoft Defender, SentinelOne, CrowdStrike Falcon and AWS Security Hub.
For teams running a single agent in a contained environment, the open-source package is complete on its own. For teams with multiple agents, shared policy requirements across business units, or existing SIEM and XDR workflows, the control plane handles the coordination that becomes impractical to do manually at scale.
Vaikora is also available on AWS Marketplace as a SaaS subscription and in the Azure Sentinel content hub as the Vaikora-AzureSecurityCenter solution, for teams that want to draw it down through existing procurement channels.
Where to start
Install the package from PyPI:
`pip install vaikora-guard-mcp`
The repository includes working examples for common agent architectures. Start by running the proxy in log-only mode against your existing agent traffic. Review what the content inspection modules flag before switching any rules to block. That gives you a real policy baseline from actual traffic rather than a configuration you wrote blind.
Once you have a week of decision telemetry, check which content categories are firing most, and whether your current block rules would have caught the top offenders. That review should shape your first set of enforcement rules before you move the proxy to production.
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