Vaikora › Blog › Product Updates
vaikora-guard-mcp on PyPI: how the proxy intercepts every tool call and what policy decisions look like
Python package, MIT license, self-hosted. That is the starting point. `vaikora-guard-mcp` is a proxy that sits inline between an MCP-connected AI agent and the tools it calls. Before any tool executes, the proxy evaluates the call against a deterministic policy and returns one of four decisions: allow, block, modify, or require-approval. Nothing executes until that decision is made.
If you want to see what that looks like in practice, install the package and read on.
What the proxy actually intercepts
When an AI agent issues a tool call over MCP, the call travels through `vaikora-guard-mcp` before it reaches the tool. The proxy extracts the tool name, the arguments, and the calling context, then runs those against your policy file. The policy evaluation is deterministic: identical input produces identical output, every time. There is no probabilistic scoring that might return different results on the same call.
The proxy ships with six content inspection modules. These check for things like credential patterns in arguments, PII in outbound payloads, and policy-defined restricted tool names or argument patterns. You configure what gets checked and what threshold triggers which decision.
Enforcement latency at P50 is 8ms. At P99 it is 45ms. The proxy handles 10,000-plus actions per second, which means it does not become a bottleneck in agentic pipelines that call tools at volume.
Installing and configuring it
```
pip install vaikora-guard-mcp
```
The package is published on PyPI. After install, you point your MCP client at the proxy endpoint instead of directly at the tool server. The proxy forwards allowed calls unchanged and returns a structured block or approval-required response for the rest.
Policy is defined in a YAML file. A minimal rule looks like this:
```yaml
rules:
- name: block-delete-tool
match:
tool: "file_delete"
action: block
reason: "File deletion requires out-of-band approval"
```
That rule catches every call to `file_delete` before it runs. The agent receives a block response with the reason string. The decision is written to the audit log.
The audit trail
Every decision, allow or block, gets written to a SHA-256 hash-chained, tamper-evident audit log. Each entry is signed into the chain, so the sequence of decisions is ordered and verifiable. If a compliance auditor or incident responder needs to reconstruct what an agent did and when, the log gives an exact record of every policy evaluation, the input, the decision, and the timestamp.
This matters for SOC 2, HIPAA, GDPR, PCI-DSS, and ISO 27001 contexts. Vaikora ships compliance presets for each of these, so you are not writing the base policy rules from scratch.
Where the open-source boundary sits
`vaikora-guard-mcp` is MIT-licensed and self-hostable. The proxy itself, the policy engine, the content inspection modules, and the audit log are all in the open-source package. You run it on your own infrastructure.
The Vaikora Control Plane is the commercial product, hosted by Data443. It adds centralized policy management across multiple agents and environments, real-time dashboards, SIEM integrations with Microsoft Sentinel, CrowdStrike Falcon, SentinelOne, and AWS Security Hub, and the ability to route require-approval decisions to a human reviewer in a defined workflow. The proxy works without the Control Plane. The Control Plane extends what you can do with the proxy at scale.
For teams running a small number of agents in a controlled environment, the open-source proxy alone may be enough. For teams running agents across multiple environments with compliance requirements and a SOC in the loop, the Control Plane is where the policy management and visibility lives.
What to check next
Start with the PyPI package page for `vaikora-guard-mcp`. The README has a working example with a local MCP server. Write one rule, run a tool call through the proxy, and read the audit log entry it produces. That sequence takes under ten minutes and gives you a concrete picture of what enforcement at the call layer looks like before you decide how far to take the configuration.
If you are running agents against sensitive data or external APIs, map your highest-risk tool names first. Those are the ones where a block or require-approval rule pays off immediately.
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