Skip to content

Getting Started

Vinkius Cloud is a managed runtime for Model Context Protocol (MCP) servers. It sits between your existing APIs and AI clients like Claude, Cursor, and VS Code — turning any REST endpoint into a governed, observable tool that AI agents can discover and call.

This means you don't need to write MCP server code, manage WebSocket infrastructure, or implement authentication flows. You provide an API; Vinkius Cloud handles the protocol translation, data protection, and global distribution.

What is MCP?

The Model Context Protocol is an open standard created by Anthropic that defines how AI clients discover and invoke external tools. When Claude asks "what tools are available?", the MCP server responds with a structured list. When Claude decides to call a tool, the server translates that into an API request. Vinkius Cloud automates this entire layer.


Sign in

Go to cloud.vinkius.com and authenticate with Google or GitHub. Authentication is passwordless — the platform uses OAuth 2.0 and creates your account on first login.

  • Google — Works with Google Workspace and personal Gmail accounts.
  • GitHub — Recommended if your team already collaborates through GitHub, since it simplifies team invites later.

Once authenticated, you land on the Dashboard.


Your command center
The Dashboard shows everything.
In real time.
Servers, tools, token usage, DLP redactions, FinOps savings, live request feeds — one page, always current. You never need to leave the Dashboard to understand the state of your infrastructure.

The Dashboard

OVERVIEW
Welcome back
Infrastructure overview
BUSINESS PLAN
21 / 500K requests
TOTAL SERVERS
5
SERVERS_COUNT
ACTIVE
5
ACTIVE_COUNT
TOTAL TOOLS
47
TOOLS_COUNT
DLP ENABLED
5
DLP_COUNT
// TOKEN ANALYTICS
86K
TOTAL TOKENS (30D)
2.8K
AVG / REQUEST
31
TOTAL REQUESTS
01pet_find_pets_by_status85K tk
02store_get_inventory71 tk
03pet_get_pet_by_id66 tk
// EGRESS FIREWALL
5SERVERS PROTECTED
● DLPFINOPS
PII redactions182
Payloads truncated6
EGRESS_STATUSACTIVE

The Dashboard is your operational hub. Everything updates in real time, so you can monitor your entire MCP infrastructure without leaving the page.

Stats grid

The top row shows four key metrics at a glance:

MetricWhat it tells you
Total ServersHow many MCP servers you have deployed across all modes (OpenAPI, Manual, Skills)
ActiveHow many are currently online and accepting connections. A gap between this and Total Servers means something needs attention.
Total ToolsThe combined count of MCP tools across all servers. If you imported a 50-endpoint OpenAPI spec, this reflects those 50 tools.
DLP EnabledHow many servers have Data Loss Prevention active. Ideally, this matches your Active count.

Token analytics

AI models charge per token — and every MCP tool call consumes tokens for the tool description, the request payload, and the response. The Token Analytics panel tracks this over a rolling 30-day window:

  • Total Tokens — Aggregate token consumption across all tool executions. This number directly correlates with your AI model costs.
  • Average per Request — A per-call breakdown that helps you identify expensive tools. If a single tool averages 5,000 tokens per call, it may benefit from FinOps Guard or response optimization.
  • Top Tools — A ranked list of your most token-intensive tools. Use this to find optimization targets — a tool consuming 85K tokens is worth investigating.

No traffic yet?

If you haven't connected an AI client, this section shows a prompt guiding you to the Quick Connect flow. Token analytics populate automatically once your first tool call is processed.

Egress Firewall

The Egress Firewall is the security layer that sits between your upstream APIs and the AI models consuming your MCP tools. It operates at the response level: every payload is inspected before it reaches the requesting model.

The panel surfaces two key indicators:

  • PII Redactions — How many sensitive fields (emails, phone numbers, credit cards) have been automatically removed from API responses. A high number here means DLP is actively protecting your data.
  • Payloads Truncated — How many oversized responses were truncated by FinOps Guard to prevent context window waste.

Live Feed

The bottom of the Dashboard streams tool executions in real time. Each entry shows the timestamp, which tool was called, the semantic verb (whether it was a read operation, a mutation, or a destructive action), latency in milliseconds, and the HTTP status code.

This feed is essential during integration: when you connect Claude Desktop and ask it to use a tool, you see the request appear here within seconds. It confirms the end-to-end connection is working and gives you immediate visibility into performance.


THREE WAYS TO DEPLOY
Spec. Code. Knowledge.
Pick your path.
01OPENAPI IMPORT
Paste a spec URL
The platform reads your OpenAPI spec, auto-discovers every endpoint, and deploys them as MCP tools. Zero code. Under 60 seconds.
02MANUAL API
Point to any endpoint
No published spec? Name your base URL, configure auth, and add tools manually. Full control over what gets exposed.
03AGENT SKILLS
Write knowledge blocks
Define procedures, playbooks, and FAQs as text. No API required. The AI discovers and uses them through progressive disclosure.

First steps

When you first create an account, an Onboarding Banner appears on the Dashboard. It walks you through three milestones:

  1. Deploy your first server — Choose from OpenAPI Import (paste a spec URL), Manual API (name your endpoint), or Agent Skills (define text-based knowledge). Most users start with OpenAPI Import because it requires nothing more than a URL.
  2. Connect an AI client — Use Quick Connect to link Claude Desktop, Cursor, VS Code, or any MCP-compatible client. You get a ready-to-paste Config JSON block.
  3. Observe your first request — Ask the AI client to use one of your tools and watch the Live Feed update in real time.

The banner disappears automatically once your server begins receiving traffic.


ARCHITECTURE
What happens when
an AI calls a tool.

The request lifecycle

When an AI client calls a tool through Vinkius Cloud, five things happen in sequence:

01
Authentication
The AI client sends an MCP tool invocation to your server's unique URL. The platform validates the connection token embedded in the path.
02
Protocol translation
The platform translates the MCP call into an HTTP request to your upstream API, injecting stored credentials (Bearer token, Basic Auth, or Custom Header) from the encrypted vault.
03
API execution
Your API processes the request and returns a response. This happens inside a V8 Isolate sandbox with no shared memory between servers.
04
Egress Firewall
The response passes through the security layer: DLP scans for PII and redacts it in-place; FinOps Guard checks payload size and truncates if needed. Credentials are never logged.
05
Delivery
The sanitized response is returned to the AI client as a structured MCP result. The entire lifecycle completes in milliseconds.

Key concepts

Before diving deeper into the platform, here are the concepts you'll encounter throughout the documentation:

ConceptWhat it means
MCP ServerA hosted endpoint that exposes tools to AI clients via the Model Context Protocol
ToolA single function that an AI model can discover and call — typically mapped to an API endpoint
Connection TokenA secret embedded in the MCP URL that authenticates AI client connections
DLPData Loss Prevention — automatically redacts PII from API responses before they reach AI models
FinOps GuardPayload size control that prevents oversized API responses from consuming excessive tokens
TOONTool Object Optimized Notation — a compression format that reduces tool description size by 30-50%
Egress FirewallThe security layer that inspects every response before forwarding it to AI clients
Agent SkillsText-based knowledge blocks (procedures, playbooks, FAQs) that AI agents can discover and use

Next steps


Frequently Asked Questions

What is Vinkius Cloud and how does it relate to MCP?

Vinkius Cloud is a managed runtime for Model Context Protocol (MCP) servers. It sits between your existing APIs and AI clients — handling protocol translation, authentication, data protection, and hosting so you don't need to build or maintain MCP infrastructure yourself.

How long does it take to set up my first MCP server?

Most users go from sign-in to a live MCP server in under 60 seconds. The fastest path is OpenAPI Import: paste a spec URL, configure authentication, and click Deploy. The server is immediately accessible to AI clients.

Which AI clients can connect to Vinkius Cloud MCP servers?

Any MCP-compatible client works, including Claude Desktop, Claude Code, Cursor, VS Code with GitHub Copilot, Windsurf, Cline, and OpenAI Codex. The platform provides a ready-to-paste Config JSON block and Cursor deep link for instant setup.

Do I need an existing API to create an MCP server?

No. While OpenAPI Import and Manual API modes require an API, the Agent Skills mode lets you create MCP servers from pure text — procedures, playbooks, FAQs, and decision trees that AI agents can discover and follow without any API backend.

What security features are included by default?

Every server automatically gets Data Loss Prevention (DLP) that redacts PII from responses, FinOps Guard that prevents token waste from oversized payloads, SSRF protection that blocks internal network access, and AES-256 credential encryption in an isolated vault.

What does the Dashboard show?

The Dashboard provides real-time visibility into your entire MCP infrastructure: server count and status, total tools deployed, token consumption analytics, DLP redaction counts, FinOps savings, and a live feed of every tool execution with timestamps, latency, and HTTP status codes.