Vaikora › Blog › Product Updates
vaikora-guard-mcp is on PyPI: install it, write a policy file, block tool calls before they run
The package is called vaikora-guard-mcp. It is MIT-licensed, written in Python, published on PyPI, and self-hostable. It sits inline between an AI agent and the MCP server it talks to, checks every tool call against a policy file before that call executes, and writes each decision to a tamper-evident audit log. That is the whole thing.
If you want to try it without reading further: `pip install vaikora-guard-mcp`.
What the proxy actually does at the call layer
When an AI agent sends a tool call, vaikora-guard-mcp intercepts the request before it reaches the MCP server. It runs the call through a set of content inspection modules, matches the request against your policy rules, and returns one of four decisions: allow, block, modify, or require-approval. The underlying LLM or tool never sees the request until a decision is made.
The decision is deterministic. Identical input produces identical output, every time. There is no probabilistic scoring that produces a different result on two runs of the same call.
Latency at P50 is 8ms. At P99 it is 45ms. The proxy handles 10,000-plus actions per second, so it does not become a bottleneck in a pipeline that calls tools at volume.
Every decision is written into a SHA-256 hash-chained audit log. Each entry is signed into the chain, so you get an ordered, tamper-evident record of what was allowed or blocked and why.
A concrete example: blocking exfiltration through a file-write tool
Say your agent has access to a `write_file` tool and a `send_webhook` tool. Without a policy gate, there is nothing stopping the agent from writing sensitive data to disk and then posting that file's contents to an external endpoint. The permissions on those tools might be scoped correctly inside your IAM layer, but that does not constrain what data moves through the calls themselves.
With vaikora-guard-mcp, you write a policy rule that blocks any `send_webhook` call whose payload matches a pattern for PII or secrets. The proxy inspects the call body before execution. If the content triggers the rule, the call is blocked and logged. The webhook never fires.
You can also set a rule that requires human approval for any `write_file` call that targets a path outside a designated working directory. The agent pauses, the approval request surfaces to a reviewer, and execution resumes only on explicit sign-off. The audit log records the approval, who granted it, and the timestamp.
Policy configuration and compliance presets
Policy rules live in a configuration file. You define conditions, the tool name or names the rule applies to, and the decision to enforce. No custom code required for the enforcement logic itself.
For teams working under specific compliance frameworks, vaikora-guard-mcp ships presets for HIPAA, PCI-DSS, GDPR, SOC 2, and ISO 27001. A preset loads a baseline set of rules appropriate to that framework. You can extend the preset with additional rules or override individual decisions.
The proxy supports 12 LLM providers with no SDK migration. You point it at your existing model API and it handles the interception layer without requiring changes to how the agent is built.
How it connects to the broader control plane
vaikora-guard-mcp is the open-source, self-hostable component. The Vaikora Control Plane, hosted by Data443, is the commercial layer that adds centralized policy management, cross-agent audit aggregation, and native integrations with Microsoft Sentinel, CrowdStrike Falcon, SentinelOne, and AWS Security Hub.
If you are running a single agent in a contained environment, the open-source proxy gives you enforcement and auditability without any external dependency. If you are operating multiple agents across teams or need to push AI agent threat signals into your existing security operations tooling, the Control Plane handles that without custom engineering.
The package is also available as a SaaS subscription on AWS Marketplace, which supports private offers and EDP draw-down if your procurement runs through AWS.
What to check next
Start with the PyPI package. Install it, run it against a test agent, and read the audit log output after a few tool calls. Check whether the decisions match what your policy file specifies. If they do not, the policy rule is the place to look, not the proxy logic.
If you are evaluating it for a compliance-sensitive environment, load the relevant preset and run a set of tool calls that should trigger blocks. Confirm that each blocked call appears in the audit chain with the correct rule reference before you deploy it in front of production agents.
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