About Vantio Gate
Vantio Gate is an MCP service for dry-run evaluation of Policy Latch decisions for AI agents. It does not enforce policies live through MCP; the project also includes Optics for recording AI egress activity.
Key features
- 01Policy Latch dry-run evaluation
- 02No live enforcement through MCP
- 03Optics records destinations and processes
- 04Export recorded connection events
Use cases
- 01Evaluate policy decisions for agents
- 02Review outbound connections from monitored processes
MCP support
The Vantio Gate MCP package evaluates Policy Latch policies in dry-run mode and does not enforce them live.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
npx -y @vantio/gate-mcpVANTIO_API_KEYSecretOptional Pro API key for gate_get_policy / gate_residual_risk against your control planeVANTIO_API_BASEOptional Pro API base URL (default https://api.vantio.ai)Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.
Tools
6Saved tool list · updated 2w ago
| Parameter | Type | Required | Description |
|---|---|---|---|
| policy | object | No | Policy object to evaluate against |
| policy.dry_run | boolean | No | |
| policy.enforce | boolean | No | |
| policy.pii_types | array | No | |
| policy.redact_pii | boolean | No | |
| policy.allowed_hosts | array | No | |
| policy.blocked_hosts | array | No | |
| policy.spend_cap_usd | number | No | |
| policy.max_request_bytes | number | No | |
| hostname | string | Yes | Destination hostname, e.g. api.openai.com |
| spent_usd | number | No | Session spend so far in USD |
| request_bytes | number | No | Request body size in bytes |
Details
From the project READMEVantio
Linux security for AI.
This repository contains Vantio Optics, the Phantom Engine installer, and the Python SDK.
Vantio Optics (Open Core)
Free, local-first observability for AI egress. Optics is the free Observe tier. It records the destination, the process, the size, and the timing of a supported outbound call. Does not block actions or retain prompts or completions.
The Optics home is moving to github.com/vantioai/vantio-optics. Publishing still runs from this repository. The editor extension stays at extensions/vantio-optics. Commands and local install notes are in extensions/vantio-optics/README.md.
Install
These are the versions npm and PyPI served on 2026-10-21. Manifests in this git tree can be newer than the registry.
| Package | Registry | Version |
|---|---|---|
@vantio/cli | npm | 0.3.24 |
@vantio/agent-sdk | npm | 0.2.4 |
vantio-agent-sdk | PyPI | 3.1.1 |
@vantio/optics-mcp | npm | 0.1.2 |
@vantio/gate-mcp | npm | 0.1.0 |
@vantio/gate-mcp is a compatibility package for an older integration.
Optics CLI, on macOS, Linux, and Windows (WSL):
npm install -g @vantio/[email protected]
vantio run node agent.js
vantio run injects the Node interceptor with NODE_OPTIONS --require for node, npx, tsx, and ts-node. On Node it covers fetch, undici, http/https, http2, net/tls, WebSocket, and spawned curl or wget.
Python SDK:
pip install vantio-agent-sdk==3.1.1
Wrap a Python callable with @shield. Prefixing vantio run python does not intercept by itself.
Node SDK:
npm install @vantio/[email protected]
The Node SDK exposes shield(). Outbound Node coverage comes from vantio run.
Optics MCP:
npm install @vantio/[email protected]
Free Optics needs no account and no API key.
Browser paths stay outside this wrap.
A process can miss the wrapper when it opens a socket the Node interceptor does not see, starts a subprocess outside vantio run, or uses a runtime other than Node without the Python SDK. Optics records the paths it wrapped.
vantio coverage prints the processes and connections recorded on this system, and which of those paths are on a local allowlist. vantio destinations flags the first time a recorded program connects to a destination that is not on that list. vantio export sends those events as OTLP/HTTP JSON, syslog or JSON lines, or a webhook. Optics does not stop the call.
Telemetry is disabled by default. Set VANTIO_TELEMETRY=1 to opt in. VANTIO_TELEMETRY_DISABLED=1 or DO_NOT_TRACK=1 override that opt-in.
Phantom Engine installer
Phantom Engine is enforcement and control for enrolled Linux systems. Contact Vantio. The product page is vantio.ai/phantom-engine.
The installer source is packages/vantio-install. vantio-install is not on npm or PyPI. Operator install uses the sealed wheel described in packages/vantio-install/docs/QUICKSTART.md. The commands are vantio-install and vantio-verify. That install does not replace @vantio/cli. The installer notes in that package target Ubuntu 24.04 LTS on x86_64.
Documentation
- Site: vantio.ai
- Docs: vantio.ai/docs
- Optics: vantio.ai/optics
- Phantom Engine: vantio.ai/phantom-engine
- CLI: packages/vantio-cli/README.md
- Python SDK: packages/vantio-agent-sdk-py/README.md
- Editor extension: extensions/vantio-optics/README.md
License
@vantio/cli, @vantio/agent-sdk, and vantio-agent-sdk declare MIT. License files for the CLI and the Python SDK are packages/vantio-cli/LICENSE and packages/vantio-agent-sdk-py/LICENSE. The Node SDK declares MIT in packages/vantio-agent-sdk/package.json. There is no root LICENSE file. The MIT grant covers the published packages named here.
Security
Report a suspected vulnerability to [email protected]. The process is SECURITY.md.
Contributing
See CONTRIBUTING.md and CODE_OF_CONDUCT.md.
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.