Vaikora › Blog › Product Updates

What vaikora-guard-mcp actually does when an MCP tool call arrives, and why we published it on PyPI

Product Updates · Nov 6, 2026 · 3 min read

Most policy enforcement for AI agents happens after the fact: logs reviewed, alerts triaged, damage already done. vaikora-guard-mcp sits inline and makes an allow, block, modify, or require-approval decision before the tool call executes. The package is MIT-licensed, written in Python, and available on PyPI right now.

What the proxy does, mechanically

When an AI agent running over MCP issues a tool call, the request hits vaikora-guard-mcp first. The proxy evaluates the call against a deterministic policy set: same input, same decision, every time. If the call matches an allow rule, it passes through. If it matches a block rule, it stops there. If a rule requires human approval, the call is held until a response arrives.

The decision is written into a SHA-256 hash-chained audit log before anything downstream sees the request. That ordering matters for compliance investigations: you get a tamper-evident record of every enforcement decision, not a log that an agent process could theoretically write to after the fact.

P50 enforcement latency is 8ms. P99 is 45ms. At 10,000+ actions per second, the gate is not a throughput bottleneck for realistic production agent workloads.

Why open source, and what that changes

Publishing under MIT shifts the trust question. A team evaluating a policy gate for production agents cannot take "trust us" as an answer. They need to read the inspection logic, verify the audit chain implementation, and confirm the proxy does not introduce its own data handling risk.

With vaikora-guard-mcp on PyPI, that review happens before deployment. Security engineers can diff the source, run it in a test environment, and confirm behavior matches documentation. That is a different audit posture than evaluating a closed binary.

The six content inspection modules ship in the open-source package. The commercial product is the Vaikora Control Plane, hosted by Data443, which adds fleet-wide policy management, compliance presets for HIPAA, PCI-DSS, GDPR, SOC 2, and ISO 27001, and native integration with Microsoft Sentinel, CrowdStrike Falcon, SentinelOne, and AWS Security Hub.

A concrete scenario

Suppose an agent is authorized to query a customer database and summarize results. Without a policy gate, nothing stops it from issuing a query that returns a full record set including fields outside the intended scope, then passing that data to a downstream tool call. The agent's intent may be benign; the data exposure is not.

With vaikora-guard-mcp in the path, you write a policy rule that blocks or modifies tool calls matching that pattern. If the agent attempts the out-of-scope query, the proxy stops it before the database sees the request. The audit log records the blocked call, the matched rule, and the timestamp. If a compliance reviewer asks what the agent did during a session, the hash-chained log answers that question without relying on the agent's own output.

What you need to get started

Installation is a single pip command. The proxy is self-hosted, so traffic does not leave your environment unless you connect to the hosted control plane. It supports 12 LLM providers with no SDK migration required, and it speaks MCP natively, so it drops into existing agent architectures without rewiring the client.

For teams that need policy management across multiple agents, the control plane adds centralized rule authoring, audit log search, and the SIEM integrations. For teams that want to evaluate the enforcement logic first, the open-source package is the right starting point.

What to check next

Pull `vaikora-guard-mcp` from PyPI and read the inspection module source before you configure anything. Verify that the audit log format fits your retention and investigation requirements. If you are running agents against regulated data, check whether the HIPAA or PCI-DSS presets in the control plane cover the specific tool categories your agents use, or whether you need custom rules. The compliance presets are a starting point, not a substitute for reviewing what your agents actually call.

More from the Vaikora blog