VINKIUS
Any API → MCP server in 30s

$ vurb deploy → LIVE!The MCP framework.

Vurb.ts introduces MVA — Model-View-Agent — where a Presenter layer controls exactly what the AI perceives. Egress firewall, FSM state gates, governance lockfile — built for production MCP servers.

Try for Free·No credit card
Open-Source TypeScript Framework

Control what
AI perceives.
Ship in
seconds.

Vurb.ts is a production-grade, open-source TypeScript framework for MCP servers. Type-safe tools, built-in auth, real-time inspector, and one-command deploy to our global edge. You write the logic — the framework handles everything else.

View Vurb.ts on GitHub

One-click deploy. Use instantly with your favorite AI clients:

ClaudeDesktop & Code
CursorIDE
ChatGPTOpenAI
WindsurfIDE
VS Code+ Copilot
ClineTerminal
What's Built In

Prompt it.
Ship it.
2 minutes.

Vurb.ts was engineered so AI masters it instantly. No learning curve. Production-safe. Prompt what you need — get a production MCP server on a battle-tested foundation.

01

Context Tree-Shaking

Without Presenter~2,000 tokens/call
With .rules() / .systemRules()~200 tokens/call
90% reductionrules travel with data

Like webpack tree-shaking removes unused code, .rules() removes unused rules from the context window. Domain rules only appear in the prompt at the precise moment the agent processes that entity.

02

SSR for Agents

SERVERcompiles deterministically
ECharts
AB
Mermaid
━━━━
Table
AGENT[SYSTEM] pass-through → forwards unchanged

The AI is a messenger, not a renderer. Charts, diagrams, and tables are compiled server-side via .ui() — 100% deterministic. Visual hallucinations drop to zero.

03

Egress Firewall

Zod .schema() — strips fields at RAM level

id string

name string

status enum

password_hash stripped

ssn stripped

internal_notes stripped

.agentLimit(50)truncates + guidance

Native Prompt Injection Defense. .schema() physically strips undeclared fields at RAM level. .agentLimit() truncates massive arrays — securing token economics.

AI-First by Design

You don't
learn it.
Your AI does.

The Fluent API is elegant for humans — but no one should have to learn another framework. Vurb.ts was built AI-First. Any coding assistant understands it in milliseconds and ships a production-grade MCP server on a solid foundation.

llms.txt included Fluent API Production-safe

Type-Safe Tools

Semantic verbs — query, mutation, action. Zod-validated params. Auto-generated schemas. Zero runtime surprises.

const listUsers = f.query('list-users')

.withString('role')

.withNumber('limit')

.returns(UserPresenter)

query mutation action

Auth & Security

JWT, API Keys, SSRF guard, rate limiting, domain allowlist. 9 governance modules. One .use() call.

Auth0.3ms
Rate Limit0.1ms
SSRF Guard0.2ms
Execute12ms
JWTAPI KeyOAuth

Presenter Layer

Strip fields. Embed rules. Suggest next actions. Render charts. The AI sees exactly what you decide — nothing more.

SCHEMARULESUI BLOCK
ECharts
Mermaid
ABC

Real-Time Inspector

Live TUI dashboard. Every call, every param, every response — in real time. Local dev and production.

vurb inspect
list-users200 OK
create-task201 OK
delete-user403 DENIED
3 calls | 14ms avg | live

Governance Lockfile

Cryptographic contract per tool. Schema drift breaks CI. SOC2-ready audit trail. Zero trust by default.

vurb.lockSHA256

- "digest": "f8e5d4c3..."

+ "digest": "9a8b7c6d..."

"actions": ["create", "list"]

9 modules CI gates drift

One-Command

Bundled, compressed, live on our global edge — under 4 seconds. Or self-host. Same code, zero rewrites.

Config Bundle Compress Edge

$ vurb deploy

Bundling with esbuild

Compressing (142KB → 38KB)

Deploying to Edge

url: https://cloud.vinkius.com/mcp

The Developer Experience

From terminal
to live.

Write code. Run one command. Live on the global edge in 4 seconds. No servers to configure, no pipelines to set up, no infrastructure headaches.

$ vurb deploy

Reading configuration

Resolving entrypoint (src/server.ts)

Bundling with esbuild

Compressing (142KB → 38KB gzip, 73% smaller)

Deploying to Edge

my-mcp-server — deployed to edge

id: dep_a8f3c2d1

size: 142KB → 38KB gzip (73% smaller)

url: https://cloud.vinkius.com/mcp/my-server

time: 3.8s

Smart Routing

ONE tool, discriminator-routed. Fewer tokens per call.

AI

Agentic HATEOAS

.suggest() / .suggestActions() with explicit hints based on data state.

> render chart
> type: bar
> data: revenue Q1-Q4
> mermaid: flow
> A → B → C
> summary: 3 items
> echarts: ready
> blocks: served
> render chart
> type: bar
> data: revenue Q1-Q4
> mermaid: flow
>render_chart()

UI Blocks

ECharts, Mermaid diagrams, summaries — server-rendered.

Start Now
$npx vurb create my-server

Describe what to build. The AI already knows every tool, presenter, and guard. Production-grade architecture lands in minutes — not weeks. One command. Live everywhere.

Open source. Free forever. Apache 2.0.