Home | Blog | Vaikora Guard MCP: Running Policy Before Every Agent Tool Call With a Self-Hosted MIT Proxy

Vaikora Guard MCP: Running Policy Before Every Agent Tool Call With a Self-Hosted MIT Proxy

One gap keeps showing up in production AI agent deployments: the tool call executes first, and any policy check happens after. By then the action is already taken, the file is already written, the API is already called. Vaikora Guard MCP is built to close that gap by sitting inline between the agent and the tool, enforcing policy before execution runs.

Here is what the library actually does, how it works technically, and why the open-source release matters for teams running MCP-based agents.

What Vaikora Guard MCP Actually Is

Vaikora Guard MCP is an MIT-licensed Python proxy published on PyPI as `vaikora-guard-mcp`. It implements the Model Context Protocol, so it intercepts every tool call an agent makes before that call reaches the underlying tool or LLM. At that interception point, it evaluates the call against a policy definition and returns one of four deterministic decisions: allow, block, modify, or require-approval.

The critical word there is deterministic. Identical input produces an identical policy decision every time. There is no probabilistic scoring or model-based judgment at the enforcement layer. The policy is a set of rules, and the rules are evaluated consistently.

Being self-hostable matters for teams with data-residency requirements or who cannot route agent traffic through a third-party service. Because the library ships under the MIT license, it can be forked, audited, extended, and embedded into existing infrastructure without licensing friction.

How the Inline Enforcement Works

When an AI agent constructs a tool call, Vaikora Guard MCP intercepts it at the MCP layer. The proxy inspects the call through its content inspection modules, evaluates it against the configured policy, and either passes it through or blocks it before the tool ever receives it.

A concrete scenario: an agent running against a customer database is configured to allow read operations on certain tables and block any write, delete, or schema-modification call. Without inline enforcement, a prompt injection or a malformed agent step could issue a destructive write. With the proxy in place, the write call hits the policy gate, matches the block rule, and never reaches the database adapter. The agent receives a policy-denied response instead.

The enforcement latency at P50 is 8ms, and at P99 is 45ms. For most agent workflows, that overhead is not meaningful. The proxy also supports 10,000+ actions per second, so it does not become a throughput bottleneck in high-volume pipelines.

Auditability Built Into the Decision Record

Every policy decision the proxy makes is written to a SHA-256 hash-chained, tamper-evident audit log. Each entry in the chain references the hash of the prior entry, so any modification to a past decision breaks the chain and is detectable.

That design matters for compliance teams. If an auditor asks what policy was applied to a specific agent action on a specific date, the log provides an ordered, tamper-evident record of the decision, the input that triggered it, and the outcome. Vaikora ships compliance presets for HIPAA, PCI-DSS, GDPR, SOC 2, and ISO 27001, so teams working under those frameworks have a starting policy configuration rather than building from scratch.

The Relationship Between the Open-Source Proxy and the Control Plane

Vaikora Guard MCP is the enforcement component. The Vaikora Control Plane, hosted by Data443, is the commercial product that sits on top of it. The control plane handles policy management across multiple agents, centralized audit log review, SIEM integrations (Microsoft Sentinel, SentinelOne, CrowdStrike Falcon, AWS Security Hub), and threat intelligence pushes.

Teams that want to run enforcement locally without the control plane can do that. The proxy enforces policy independently. Teams that need centralized governance across dozens of agents, incident escalation into existing security tooling, or cross-agent audit trails will want the control plane connected.

Vaikora is available on AWS Marketplace as a SaaS subscription and ships in the Azure Sentinel content hub as the Vaikora-AzureSecurityCenter solution, so it fits into standard enterprise procurement and deployment patterns.

What to Check Next

If you are running MCP-based agents in production or staging, the first thing to verify is whether any policy gate exists between the agent and its tools. If not, start with the open-source library. Install `vaikora-guard-mcp` from PyPI, review the default policy configuration, and run a tool call through it to confirm the interception and decision logging are working as expected before you define production rules. The MIT license means you can inspect every line of the enforcement logic before you rely on it.