Vaikora › Blog › Product Updates

vaikora-guard-mcp: the MIT-licensed MCP proxy that blocks tool calls before they execute

Product Updates · Nov 10, 2026 · 3 min read

An AI agent can call a file-write tool, a database query, or an external API in milliseconds. Without a gate between the agent and those tools, the only control you have is after the fact. vaikora-guard-mcp puts a policy check inline, before each tool call executes, and it is open source under the MIT license.

The package is published on PyPI as `vaikora-guard-mcp`. Install it, configure a policy, point your MCP-compatible agent at it, and every tool call goes through a decision before it reaches the target.

What the proxy actually does

vaikora-guard-mcp sits between your AI agent and the LLM or tools it calls. When the agent issues a tool call, the proxy intercepts it, evaluates the call against your defined policy, and returns one of four decisions: allow, block, modify, or require-approval. The tool does not execute until a decision is made.

The proxy is written in Python and supports MCP natively, so no SDK migration is required. It connects to the major hosted model APIs through built-in adapters. If you are already running an MCP-compatible agent, the integration point is the proxy URL, not your agent code.

Every decision is written to a SHA-256 hash-chained audit log. Each entry is signed into an ordered chain, so the record is tamper-evident. That matters for compliance reviews, incident investigation, and any situation where you need to show exactly what an agent did or tried to do, and what policy decided.

A concrete example

Say you have an agent that processes customer support tickets and can call a `send_email` tool. Without a policy gate, the agent calls that tool whenever its reasoning says to. With vaikora-guard-mcp in front of it, you write a policy rule that flags outbound email containing patterns associated with PII or account credentials. Any call matching that rule gets a block or a require-approval decision. The email does not send. The decision lands in the audit log with the full context of the call.

That is not detection after the fact. The tool call never executed.

Self-hosted by default

The proxy runs in your own environment. There is no external service dependency for the enforcement path itself. The enterprise control plane (hosted by Data443) adds a UI for policy management, aggregate telemetry, and the SIEM integrations, but the core enforcement runs wherever you deploy the proxy.

For teams that need compliance presets, Vaikora ships configurations for SOC 2, HIPAA, GDPR, PCI DSS, and ISO 27001. Those map to concrete policy rules rather than generic checkboxes. You can apply a preset and extend it, or write policies from scratch.

The proxy also integrates natively with Microsoft Sentinel, CrowdStrike Falcon, SentinelOne, and AWS Security Hub. If an agent action triggers a high-severity policy decision, that signal can be pushed into your existing security operations tooling without custom engineering.

Who this is for

Security engineers who want a policy gate on agent tool calls without waiting for a vendor to build it into their platform. AI platform teams who need auditability for every action an agent takes. Compliance teams who need a record of what agents did and what policy governed them.

The MIT license means you can read the code, modify it, and run it in any environment. The PyPI package means setup is `pip install vaikora-guard-mcp`.

What to check next

Start with the PyPI package page to review the install instructions and default configuration. Look at which tools your existing agents can call, and think about which of those calls should require a policy decision before they execute. That list is your starting point for writing your first policy rules.

If you are already running a SIEM, check whether the native integrations cover your stack before you build a custom log pipeline.

More from the Vaikora blog