Remote MCP server · OAuth sign-in · Hosted in Germany (GDPR)

Connect Claude, ChatGPT or Cursor to the AI app builder over MCP

Chattee MCP server is a remote Model Context Protocol server that lets AI assistants such as Claude Code, Claude.ai, Cursor, Codex and ChatGPT create, change, deploy and publish full web apps on Chattee.

How do I connect my assistant to Chattee?

Pick your assistant below and copy one line or one URL. Claude, ChatGPT, Cursor and Claude Desktop sign in with your Chattee account, so there is no key to paste; Claude Code is one command plus a browser sign-in. Codex, scripts and the Agent SDK use a personal access token. It takes about two minutes.

No token needed: you sign in with your Chattee account

Add Chattee once for your user, so it is there in every project, then sign in. Inside Claude Code, /mcp signs you in too.

1 · Add the server

claude mcp add --transport http chattee --scope user https://chattee.ai/mcp/connect

2 · Sign in

claude mcp login chattee
Sharing the setup with a team? Use a .mcp.json

Commit this file at the repository root. It holds no token: Claude Code reads each person’s own from the CHATTEE_PAT environment variable.

.mcp.json at the repository root

{
  "mcpServers": {
    "chattee": {
      "type": "http",
      "url": "https://chattee.ai/mcp/connect",
      "headers": {
        "Authorization": "Bearer ${CHATTEE_PAT}"
      }
    }
  }
}

The snippet reads your token from CHATTEE_PAT. Create one, and the AI assistants page shows this setup again with your token filled in, once.

Create a token No account yet? Start free

Get the Chattee skill (SKILL.md) Save it as ~/.claude/skills/chattee/SKILL.md, and Claude Code knows when and how to use Chattee.

Full Claude Code setup in the docs

Server URL https://chattee.ai/mcp/connect

Start free Full setup guide

What can your AI assistant do with Chattee?

Everything the builder does, from your own chat: your assistant creates a project from a brief, relays the plan Chattee proposes, watches the build, hands you the Draft preview, publishes to a Live URL, starts from templates, edits files, sets configuration and reads the database. Full web apps with a database, auth and file storage, not front-end demos.

Create and change apps

A brief becomes a plan, a plan becomes a deployed app. Follow-ups go into the same chat, in plan or edit mode.

Approve plans

Every plan is relayed to you in full. Your assistant approves only after you said yes, or sends your feedback back.

Publish to a live URL

Draft to Live in one call: a release is built, data is copied on the first publish, and the Live URL comes back.

Templates

Start a project from a template, or freeze a finished app as one and publish it for everyone.

Files and configuration

Read and write files, set environment variables, add a database, auth, storage or email as managed resources.

Diagnostics

Run logs, deploy logs, container output, release history and your credit balance, when something needs a look.

How it works

Three steps, with you in the loop at each decision: you describe the app to your assistant, Chattee plans it and you approve the plan in your chat, then Chattee builds, deploys and hosts it and your assistant hands you the preview. Plans and money decisions stay with you; the assistant only relays them.

  1. Describe the app

    In Claude, Cursor, Codex or ChatGPT: what it does, for whom, what it must have. Attach a spec or a screenshot if you like.

  2. Review the plan in your chat

    Chattee answers with a plan: modules, database, auth, pages. Your assistant shows it in full; you approve or ask for changes.

  3. Get the live app

    Chattee builds it in 5–90 minutes and checks it in a browser; your assistant returns the Draft preview. Say publish and you get the Live URL.

  1. You:

    Build a student portal for my tutoring school, Brightpath Tutoring. Students sign in and see their own grades and attendance, and they upload their assignments there. When a student submits an assignment, I want to get an email about it. Publish it when it works.

  2. Chattee tools called:
  3. Claude:

    I've started your Brightpath Tutoring student portal. Before it can write the plan, the Chattee planner needs your answers to two questions: how students get accounts, and which look fits Brightpath Tutoring.

    • Admin invites students
    • Open self sign-up
    • Both
    • Calm academic blue
    • Warm & encouraging
    • Fresh growth green
  4. You:

    Open self sign-up for students. For look and feel, go with calm academic blue (navy blue and white). Notification emails should go to anna@brightpath.example.

  5. Chattee tools called:
  6. Claude:

    I passed on your answers: open self sign-up, and calm academic blue in navy and white. Chattee has written its plan and is waiting for your approval. Nothing has been built yet.

    Plan from Chattee, waiting for your approval

    Modules

    • Node.js back end
    • React front end

    Managed resources, provisioned on approval

    • Postgres database
    • Sign-in
    • Email
    • File storage
  7. You:

    This plan looks great and matches what I wanted. For the admin email list, please use anna@brightpath.example for now; I'll add the other two teachers' emails myself later. Approved, go ahead and build it.

  8. Chattee tools called:
  9. Claude:

    The Brightpath Tutoring student portal is built and live: https://brightpath-tutoring.chattee.ai. Students sign up, see only their own grades and attendance, and upload their assignments; every submission emails you. All automated tests passed. The build ran past its time limit once, and I resumed it within the allowance it already had. Emails go out once you connect Brevo, Mailchimp or Amazon SES in the Data tab.

  10. You:

    Thanks for getting it live! I'll connect an email provider in the Data tab myself.

Recorded in one of our test runs on October 1, 2026, shortened. The user in it is simulated, and the app’s address is written as it would read on chattee.ai. Make a student portal app with AI

Vibe coding that ships

Vibe coding means describing what you want and letting the AI write the code. The first draft was never the hard part; hosting, a database, sign-in, email, a domain and publishing safely are. Chattee provisions all of it for every app your assistant builds, so what you vibe code over MCP is a real app from day one.

Hosted in Germany, GDPR

Apps, databases and files run on servers in Germany, under the GDPR.

Hosting with HTTPS

Every app gets a Draft preview and a Live URL on chattee.ai, with HTTPS, from the first build.

Managed Postgres

A database per project, operated and backed up by Chattee. Your assistant can query it over MCP.

Sign-in and user accounts

Sign-up, login, password reset and magic links. Your users are a table your app can JOIN.

File storage

Uploads that survive every rebuild. Public files get a URL; private files never do.

Email

Transactional email through your own Brevo, Mailchimp or SES account. Its key never enters the app.

Your own domain

Put the Live app on your own domain when you are ready.

Draft and Live

Build on Draft, publish to Live, go back to an earlier release. Live has its own database.

Encrypted environment variables

API keys and settings, encrypted, set per environment.

Start with the free tools for a spec, a schema or a cost estimate, hand the brief to your assistant, and read what vibe coding is if the term is new to you.

How Chattee over MCP compares with other AI app builders

Coding agents write good code but leave hosting, the database, sign-in and email to you, and most AI app builders are made to be used from their own chat window. Chattee over MCP joins the two: your own assistant drives a builder that provisions that infrastructure, deploys and hosts the app, and builds only a plan you approved.

Chattee over MCP compared with a coding agent alone, a coding agent with a stack you assemble, and app builders used from their own chat
Feature Chattee over MCP Your coding agent alone Coding agent + a stack you assemble App builders in their own chat
Start from Claude, ChatGPT, Cursor or Codex Yes Yes Yes Partly some add a connector
Hosting with HTTPS, live in one call Yes No Partly connect a host Yes
Postgres provisioned per project Yes No Partly a separate account Yes usually
Sign-in and user accounts Yes No Partly a separate provider Yes usually
File storage Yes No Partly a separate service Partly varies
Email through your own provider, its key never in the app Yes No Partly you wire it up Partly varies
Draft and Live environments with rollback Yes No Partly previews; the data is on you Partly varies
Your own domain Yes No Yes through your host Yes
A plan you approve before anything is built Yes every run Partly if you ask Partly if you ask Partly varies
Checked in a real browser before you see it Yes Partly extra tooling Partly extra tooling Partly varies
Hosted in Germany, under the GDPR Yes Partly wherever you deploy Partly a region per service Partly varies
Accounts and bills to run the app One Nothing runs yet 3–5 One
  • Included
  • Partly, or with extra work
  • Not included

Pricing and limits

Connecting costs nothing. Building spends the credits of your Chattee plan, the same as in the browser, and the free plan includes credits to start. One run per project at a time, 120 tool calls per minute per user, and a run that reaches its budget pauses; your assistant raises it only to a total you name.

See the plans

Frequently asked questions

Can ChatGPT build me an app?

Yes. Turn on Developer mode in ChatGPT, add Chattee as a connection under ChatGPT Plugins, sign in, and describe the app. ChatGPT relays Chattee’s plan, you approve it, Chattee builds, deploys and hosts the app, and ChatGPT hands you the preview and the Live URL.

Do I need a paid ChatGPT plan?

Yes. ChatGPT connects to an MCP server like Chattee through its Developer mode, which OpenAI offers on its paid plans (Plus, Pro, Business, Enterprise and Edu) and on the web only; in a Business, Enterprise or Edu workspace an admin may have to allow it first. Connecting Chattee costs nothing; building spends the credits of your Chattee plan.

Which AI assistants can use Chattee?

Claude Code, Claude.ai, Claude Desktop and the Claude mobile apps, Cursor, Codex CLI, ChatGPT, and any agent built on the Claude Agent SDK or another MCP client that speaks Streamable HTTP.

Does Chattee work with VS Code, Gemini CLI or Windsurf?

It should: each of them supports remote MCP servers over Streamable HTTP, which is what Chattee is, with OAuth 2.1 sign-in or a personal access token. We have not tested them ourselves, so the docs cover the clients above; point yours at https://chattee.ai/mcp/connect and sign in, or send a token as a Bearer header.

What is an MCP server, in plain words?

A service an AI assistant can call, with a fixed set of actions it understands. The Model Context Protocol is the shared language, so one server works with every assistant that speaks it. Chattee’s server offers 35 actions: create a project, approve a plan, publish, and so on.

What is a Claude connector, and is it the same as an MCP server?

A connector is how Claude.ai, Claude Desktop and the Claude apps reach an MCP server: you add the server’s URL as a custom connector, sign in once, and its tools appear in your chats. Chattee’s connector is its MCP server at https://chattee.ai/mcp/connect; Claude Code adds the same server with one command.

Do I need to install anything?

No. The server runs on chattee.ai. You add its URL to your client, or run one command in Claude Code or Codex, and sign in.

Is the connection secure?

Yes. Clients sign in with OAuth 2.1 and PKCE or use a scoped personal access token; tokens are stored hashed and can be revoked on the AI assistants page at any time. Every call is checked against its scope and passes the same guardrails as the builder.

Where are my apps and data hosted?

In Germany. Chattee runs your apps, their databases and their uploaded files on servers in Germany and processes your data under the GDPR. Your conversation itself stays with your assistant’s provider under its own terms; Chattee receives only the tool calls.

What does it cost?

Connecting is free. Building spends the credits of your plan, exactly as in the browser; the free plan includes credits to start. Your assistant never buys credits, and raises a paused run's budget only to a total you name. A pause nobody decides on ends by itself after 30 minutes, keeping the work.

Can it deploy the app?

Yes. Every approved plan is built and deployed to a Draft preview on chattee.ai, and publishing puts it on a Live URL or your own domain, with the database, auth and file storage that come with it.

Does it work in German, Spanish, French or Portuguese?

Yes. The builder and this page exist in all four, and you can describe your app in any of them. The MCP server’s own answers are in English; your assistant relays them to you in your language.

Connect your assistant to Chattee

Two minutes of setup, then your assistant builds, deploys and publishes on Chattee, and you approve every plan.

Choose your assistant Start free