MCP 服务器

OBTO

OBTO gives your AI assistant a hosted app platform: it scaffolds, deploys, and operates real full-stack apps — pages, APIs, database, scheduled jobs — in a workspace created automatically at your first Google sign-in.

README

<h1 align="center">OBTO — The Glass Box AI Platform</h1>

<p align="center"> <strong>The first MCP-native platform where AI agents deploy production applications.</strong><br/> Describe what you want. The AI builds it and deploys it. You get a live URL. </p>

<p align="center"> <a href="https://www.obto.co">Website</a> · <a href="https://docs.obto.co">Docs</a> · <a href="https://www.obto.co/site/ob/terms">Terms</a> · <a href="https://www.obto.co/site/ob/privacy">Privacy</a> </p>

What is OBTO?

OBTO is a full-stack deployment platform built for the age of AI agents. Connect it to Claude, GPT, Codex, or any MCP-compatible model. Describe what you want in plain language. The AI writes the code, provisions the database, and deploys your application to a live URL — all in the same conversation.

No code editor. No terminal. No hosting setup. Full-stack. Not a wrapper over Vercel and Supabase — real backend, real infrastructure, one platform.

OBTO has been running in production since 2019.

Quick Start

1. Sign Up

Go to obto.co and click Sign in. Authenticate with Google. Your OBTO instance is created automatically.

2. Copy Your MCP Endpoint

After signing in, you'll see your personal MCP server URL on the landing page. Copy it.

3. Teach Your AI Client — plugin or skill file

Install the plugin (recommended where supported). This repository is itself the plugin marketplace — one URL, nothing else to configure:

Claude Code / Claude Cowork

/plugin marketplace add obto-inc/platform
/plugin install obto@obto

Installs the OBTO MCP server connection plus six task skills (build loop, deploys & large files, media upload, memory, MCP extension, troubleshooting). Details: plugins/obto.

OpenAI Codex — clone this repository, then codex plugin marketplace add "/path/to/platform" and install OBTO. The Codex package is plugins/obto-codex.

Use the skill files directly (every other harness). The skills are plain-markdown SKILL.md folders — portable to any agent harness that reads skills or instruction files:

Client	What to use
Claude Desktop	Add `SKILL.md` to your project's knowledge or attach as a file
Cursor	Place `SKILL.md` in your workspace root or `.cursor/skills/`
VS Code (Copilot)	Place `SKILL.md` in your workspace root
AntiGravity	Place `SKILL.md` in your workspace's `.agent/skills/` directory
OpenCode & other agent harnesses	Copy the `plugins/obto/skills/` folders into your harness's skills directory, or use the single-file `SKILL.md`

4. Connect to Your AI Client

Add your MCP server URL to your AI client's MCP configuration:

Claude Desktop — Go to Settings → MCP Servers → Add your OBTO endpoint URL. Claude will redirect you to authenticate via OAuth. Approve it.

Claude Web — In a project, add the MCP server URL. Authorize when prompted.

Cursor / VS Code — Add the MCP server configuration to your settings.

5. Build

Type what you want. The AI handles the rest.

"Build me a landing page for my coffee shop with a menu, gallery, and contact form."

Your app will be live at yourapp.obto.co.

What Does the Skill File Do?

The SKILL.md file is a set of instructions that teaches AI models how to build and deploy applications on OBTO. It covers:

Platform architecture — How OBTO stores code as database artifacts (no filesystem)
Deployment order — Backend first, frontend last
Collection types — Pages, JavaScript modules, stylesheets, routes, server scripts
The xe. pattern — How backend services call each other
MCP tool routing — Which tool to use for each operation
Full-stack examples — Complete contact form deployment walkthrough

Without this skill file, the AI won't know OBTO's conventions and will produce code that doesn't deploy correctly. With it, the AI becomes an autonomous developer on your OBTO instance.

Where your client supports plugins (Claude Code, Claude Cowork, Codex), prefer the plugin — the same doctrine split into six auto-loading task skills, with the MCP connection preconfigured.

Supported AI Clients

Client	MCP Support	Status
Claude Code / Claude Cowork	✅ Native plugin	Fully supported (plugin + skills)
Claude Web	✅ Native	Fully supported
Claude Desktop	✅ Native	Fully supported
Cursor	✅ Native	Fully supported
VS Code (GitHub Copilot)	✅ Native	Supported
AntiGravity	✅ Native	Fully supported
OpenAI Codex	✅ Plugin via MCP	Supported
OpenCode & other skill-compatible harnesses	✅ Via MCP + skill files	Supported