VaikoraVaikora

Vaikora › Blog › Product Updates

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

Product Updates · Nov 17, 2026 · 3 min read

Twelve lines of YAML and a pip install are enough to put a policy gate in front of every tool call an AI agent makes. That is what vaikora-guard-mcp does, and it is published on PyPI under the MIT licence so you can read the source, self-host it, and modify it without restriction.

What problem this solves

MCP (Model Context Protocol) lets AI agents call external tools: read a file, query a database, POST to an API. The protocol itself has no enforcement layer. By default, if the agent decides to call a tool, the tool runs. The only question is what happens after.

That ordering is the problem. A block that fires after execution is not a block, it is a log entry. vaikora-guard-mcp sits inline between the agent and the tool. Every call passes through the proxy, policy runs deterministically, and the tool either gets the call or it does not. The agent never touches the target system if policy says no.

What "deterministic" actually means here

Deterministic policy means identical input always produces identical output. There is no model deciding whether a call looks suspicious. The proxy evaluates the call against a rule set and returns allow, block, modify, or require-approval. Same call, same answer, every time. That property matters for audit: you can replay any decision and get the same result.

Every decision is written to a SHA-256 hash-chained audit log. Each entry is signed into the chain, so any post-hoc modification breaks the chain at the point of tampering. If a compliance review or incident investigation asks what the agent did and when, the log is the answer.

Installing and running it

```
pip install vaikora-guard-mcp
vaikora-guard-mcp --config policy.yaml
```

The proxy starts and exposes an MCP-compatible endpoint. Point your agent's tool calls at that endpoint instead of the tool directly. No SDK migration is required. vaikora-guard-mcp supports the major hosted model APIs and ships six content inspection modules covering prompt injection patterns, PII, credential leakage, and similar categories.

A minimal policy block looks like this:

```yaml
rules:
- name: block-credential-exfiltration
match:
tool: "*"
content_contains: ["api_key", "password", "secret"]
action: block
reason: "Credential pattern detected in tool call"
```

The proxy evaluates that rule before the tool call leaves the agent runtime. If the pattern matches, the call stops there.

Enforcement performance

Enforcement latency at P50 is 8ms and at P99 is 45ms. At 10,000 or more actions per second, the proxy is not a throughput bottleneck for realistic agent workloads. Detection accuracy is 99.9% in testing across the content inspection modules.

What you can build on top of it

The open-source proxy handles the runtime gate. If you need centralized policy management across multiple agents, audit log aggregation, or the compliance presets for HIPAA, PCI-DSS, GDPR, SOC 2, and ISO 27001, those are part of the Vaikora Control Plane, which Data443 hosts. The two are designed to work together, but the proxy runs independently.

The proxy also integrates natively with Microsoft Sentinel, SentinelOne, and CrowdStrike Falcon, so agent threat signals can flow into your existing security operations tooling without custom connectors.

What to check next

Pull the package from PyPI (`vaikora-guard-mcp`), review the source on GitHub, and run it against a test agent. The first thing worth checking is whether your existing tool call patterns would trigger any of the default content inspection modules. That gives you a realistic view of your current exposure before you write production policy rules.

If you are running multiple agents in parallel or need the audit chain integrated into a SIEM, look at the control plane documentation to understand what the hosted layer adds beyond what the open-source proxy provides on its own.

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

More from the Vaikora blog