Vibepush

Stop writing YAML. Start shipping.

MCP-native deployment platform. Any AI agent can deploy, monitor, and fix your apps on your own servers.

License: AGPL-3.0 Status: Validating

---

We're in early validation stage. Star this repo if you want VibePush to exist. Your star = vote that this is worth building.

---

What is VibePush?

VibePush turns your VPS into an AI-managed deployment platform. Your server speaks MCP (Model Context Protocol) — so Claude, GPT, Gemini, or any MCP-compatible AI can:

• Deploy your apps from natural language
• Monitor logs and metrics
• Fix issues automatically

No YAML. No Kubernetes expertise. Just talk to your AI.

you: Deploy ./my-app with PostgreSQL
AI:  Building image... Deployed at my-app.example.com

you: Why is it crashing?
AI:  Found OOM error. Increasing memory to 512Mi... Fixed.

---

The Problem

We talked to developers. Here's what they said:

Kubernetes is hard

"Managing YAML is a weird Sisyphean job. It is by far my least favourite language."

70% of K8s users say complexity is their top challenge. The learning curve feels like scaling Everest.

Cloud PaaS = Lock-in + Surprise Bills

Heroku outages every other Friday. Vercel surprise bills ($95k for Cara). You don't own your infrastructure.

Self-hosted tools still need DevOps skills

Coolify, CapRover, Dokku — they give you control, but when things break, you're alone with Stack Overflow.

Vibe coding deployment gap

Lovable and Bolt generate code, but where do you deploy? Their hosting = lock-in. Your VPS = complexity. Shuttle just raised $6M to solve this exact problem.

---

Our Solution

Problem	VibePush
YAML complexity	AI generates config for you
K8s learning curve	Zero knowledge required
Cloud lock-in	Your servers, your rules
Self-hosted complexity	AI troubleshooting, not Stack Overflow
Vibe coding deployment gap	Export code → deploy with one command

---

How It Works

┌─────────────────┐      MCP       ┌─────────────────┐
│   Any AI Agent  │◄──────────────►│    VibePush     │
│                 │    Protocol    │   MCP Server    │
│  Claude / GPT   │                │                 │
│ Gemini / LLaMA  │                └────────┬────────┘
│    Cursor       │                         │
└─────────────────┘                         ▼
                                  ┌─────────────────┐
                                  │   Your Server   │
                                  │      (K3s)      │
                                  │                 │
                                  │  ┌───────────┐  │
                                  │  │  Your App │  │
                                  │  │  Postgres │  │
                                  │  │   Redis   │  │
                                  │  └───────────┘  │
                                  └─────────────────┘

1. Install VibePush on your VPS
2. Connect your AI via MCP protocol
3. Talk to deploy — AI handles Docker, Helm, K3s

---

Supported AI Agents

VibePush works with any MCP-compatible AI:

Agent	Support
Claude (Anthropic)	Primary
ChatGPT (OpenAI)	Planned
Gemini (Google)	Planned
LLaMA (Meta)	Planned
Cursor	Planned
Your custom agent	Planned

One protocol. Any AI. Your infrastructure.

---

Features (Planned)

• MCP Protocol Native — Works with any MCP-compatible AI agent
• AI Deployment — Understands your code, builds images, generates Helm charts
• AI Troubleshooting — Reads logs, identifies issues, fixes automatically
• Any Infrastructure — Hetzner, Contabo, DigitalOcean, bare metal
• One-Click Databases — PostgreSQL, MySQL, Redis, MongoDB
• Auto SSL & Security — Let's Encrypt, UFW, fail2ban

---

Start Small, Scale Big

Start with one server. Add more when you need them. No migration required.

    Day 1                    Month 3                   Scale
┌─────────────┐          ┌─────────────┐          ┌─────────────┐
│  1 Server   │          │  Control    │          │  Control    │
│ ─────────── │    →     │ ─────────── │    →     │ ─────────── │
│ All-in-One  │          │  Worker 1   │          │  Worker 1   │
│ (apps+dbs)  │          │             │          │  Worker 2   │
│             │          │             │          │  Worker N   │
└─────────────┘          └─────────────┘          └─────────────┘

Example conversation:

# Running low on RAM...
you: My server is getting slow. I have a new VPS at 95.216.x.x
AI:  Adding as worker node... Done.
     Cluster now has 2 nodes, 8GB RAM total.

# Black Friday is coming...
you: Add 3 more servers for the traffic spike
AI:  Adding workers... Done. 5 nodes ready.
     Scaling my-app to 5 replicas across all nodes.

Mix any providers:

• Hetzner (Germany) + Contabo (Germany) + DigitalOcean (NYC)
• Bare metal + Cloud VPS
• ARM + x86

---

Project Status

Early Stage — Validating the Idea

We're building this in public. Before writing code, we want to validate that developers actually need this.

What We Have

• Landing page: vibepush.dev
• Pain points research
• Architecture design
• MVP implementation

How You Can Help

Action	What it means
Star this repo	"Yes, I want this to exist"
Open an issue	Share your pain points
Watch	Get updates on progress
Contribute	PRs welcome

---

Comparison

Feature	VibePush	Coolify	Railway	Heroku
AI Deployment	Any MCP AI	—	—	—
AI Troubleshooting	Any MCP AI	—	—	—
Self-Hosted	Yes	Yes	No	No
Your Infrastructure	Yes	Yes	No	No
MCP Protocol	Native	—	—	—
Multi-Node Scaling	Yes	Partial	—	—
No K8s Knowledge	Yes	Partial	Yes	Yes

---

Tech Stack (Planned)

Component	Technology
Runtime	K3s (lightweight Kubernetes)
Protocol	MCP (Model Context Protocol)
Build	Buildpacks / Docker
Ingress	Traefik
SSL	Let's Encrypt / cert-manager
Agent	Go
MCP Server	TypeScript

---

FAQ

Why MCP and not REST API?

MCP (Model Context Protocol) is Anthropic's standard for AI-to-tool communication. It's becoming the lingua franca for AI agents. By building MCP-native, VibePush works with any AI that adopts the protocol — not just Claude.

Why K3s and not Docker Compose?

K3s gives us Kubernetes power (rolling deploys, auto-healing, scaling) without Kubernetes complexity. The AI handles the complexity — you just deploy.

How is this different from Coolify?

Coolify is UI-first. VibePush is AI-first. When your app crashes at 3am, you ask your AI to fix it — not navigate a dashboard.

Will this work with ChatGPT / Gemini?

Yes, once they support MCP. We're building for the protocol, not a specific AI.

Can I start with one server and add more later?

Absolutely. This is the expected path:

1. Day 1: Single $5 VPS runs everything (control plane + apps + databases)
2. When you grow: Add worker nodes with a simple "Add server X.X.X.X" command
3. Scale: Mix providers (Hetzner + Contabo + DO), mix architectures (ARM + x86)

No migration, no downtime, no reconfiguration.

---

Community

• Website
• GitHub Issues — Feedback & feature requests

---

License

AGPL-3.0

Free to use, modify, and distribute. If you run VibePush as a service, you must open-source your modifications.

---

Star this repo if you want VibePush to exist.

Your star = validation that we should build this.

vibepush.dev