Vaikora › Blog › Product Updates
vaikora-guard-mcp is MIT-licensed and on PyPI: here is what that means for your MCP deployment
Shipping policy enforcement as open-source changes the trust model. You can read every line before you deploy it, run it in your own environment, and audit exactly what happens to a tool call before the underlying function executes.
vaikora-guard-mcp is the open-source component of Vaikora. It is written in Python, published on PyPI under the MIT licence, and self-hostable. The commercial product is the Vaikora Control Plane, hosted by Data443. The proxy itself is yours to inspect, fork, and run.
What the proxy actually intercepts
The MCP proxy sits inline between an AI agent and the tools it can call. When an agent issues a tool call, the request hits the proxy first. The proxy evaluates the call against a policy definition and returns one of four deterministic outcomes: allow, block, modify, or require-approval. The underlying function does not execute until that decision resolves.
That ordering matters. Policy runs before execution, not after. There is no race between the enforcement layer and the action.
A concrete example: an agent is asked to retrieve a customer record and forward it to an external endpoint. The tool call carries a destination URL. The proxy checks the URL against a policy rule that blocks outbound data transfers to non-allowlisted domains. The block decision fires before the API call leaves the system. Nothing is exfiltrated. The decision is written to the audit log.
How the policy check works
Policy is deterministic. Identical input always produces an identical decision. There is no probabilistic scoring or model-based inference at the enforcement layer. This is by design: a guardrail that sometimes allows the same call is not a guardrail.
The proxy ships six content inspection modules. These cover the payload, the destination, the tool name, the calling agent identity, and other attributes of the call. Policy rules reference these attributes. You write a rule, the proxy evaluates it, the outcome is binary at the boundary: the call either proceeds or it does not.
Enforcement latency runs at P50 8ms and P99 45ms. At 10,000+ actions per second, the proxy does not become the bottleneck in a high-throughput agent workflow.
The audit chain
Every decision the proxy makes is written to a SHA-256 hash-chained, tamper-evident audit log. Each entry is signed into the chain, so the log is ordered and verifiable. If a decision entry is altered or deleted, the chain breaks.
This matters for compliance. When an auditor asks what an agent did and why a specific tool call was blocked, the log answers that question with a verifiable record, not a reconstructed best guess. Vaikora ships compliance presets for SOC 2, HIPAA, GDPR, PCI-DSS, and ISO 27001, so the policy rules that produce those log entries align to the controls your auditors expect.
Running it in your environment
Because vaikora-guard-mcp is self-hostable, your tool call traffic does not leave your infrastructure at the enforcement layer. Install from PyPI, configure your policy file, point your MCP-compatible agent at the proxy endpoint instead of the tool endpoint directly.
```
pip install vaikora-guard-mcp
```
The proxy adapts to 12 LLM providers with no SDK migration. If you are already running an MCP-compatible agent setup, adding the proxy is a configuration change, not a re-architecture.
For teams that want centralized policy management, cross-agent visibility, and integration with platforms like Microsoft Sentinel, SentinelOne, or CrowdStrike Falcon, the Control Plane connects to the same proxy. The open-source layer handles enforcement; the Control Plane handles governance at scale.
What to check before you deploy
Before standing up any inline proxy in an agent workflow, verify three things.
- Latency budget. Confirm your agent's timeout settings accommodate the proxy round-trip. At P99 45ms, most workflows absorb this without adjustment, but check against your specific pipeline.
- Policy coverage. Map the tools your agents can call to the inspection modules the proxy ships. Any tool that touches sensitive data, external endpoints, or privileged operations should have an explicit rule, not a default-allow fallthrough.
- Audit log retention. Decide where the hash-chained log writes and how long you retain it. For HIPAA and PCI-DSS environments, retention requirements are defined. Set the destination before you go live.
The source code is on PyPI as `vaikora-guard-mcp`. Read the inspection module implementations before you configure policy. The enforcement logic is exactly what you read, nothing more.
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