Model Context Protocol

The MCP server thatpublishes real websites

Connect Claude or ChatGPT to Dplooy with one URL. Ask for a website and your AI builds it, publishes it to a live address, and comes back to edit it whenever you ask — with working forms, a booking widget, editable content and a chatbot.

No install
No API key
Free plan works
20+
tools your AI can use
1
URL to connect, no install
0
API keys to copy or paste
Free
on every plan, including free

Your AI can write a website. It can't publish one.

Ask any assistant for a website and you get code in a chat window. You still have to copy it, find a host, upload it, wire up a form backend, and do it all again for every change.

Without MCP

  • AI writes code into the chat
  • You copy, paste and find a host
  • Forms need a third-party service
  • Bookings need a separate SaaS
  • Every edit means rebuilding from scratch

With Dplooy MCP

  • AI publishes straight to a live URL
  • Free SSL, global CDN, custom subdomain
  • Real form endpoint with an inbox
  • Booking widget with your approval
  • Edits change only what you asked for

Two ways to connect

Same tools either way. Pick the one that matches where you chat.

Remote — claude.ai & ChatGPT

Nothing to install · recommended

Settings → Connectors
Add custom connector
Server URL
https://api.dplooy.com/mcp
Cancel
Add
  1. 1Create a free Dplooy account
  2. 2Settings → Connectors → Add custom connector
  3. 3Paste the URL above
  4. 4Sign in to Dplooy and approve — no key to copy
  5. 5Ask for a website

Custom connectors on claude.ai require a paid Claude plan; Team and Enterprise workspaces may need an admin to enable them. In ChatGPT, connectors in developer mode expose the full tool set.

Local — Claude Code, Cursor & more

Runs on your machine, reads your disk

terminal
$ claude mcp add -t stdio \
    -e DPLOOY_API_KEY=dpk_live_xxx \
    dplooy -- npx -y @dplooy/mcp-server
✓ Added MCP server "dplooy"

For Claude Desktop, Cursor or Windsurf, add the same thing to your MCP config file:

{
  "mcpServers": {
    "dplooy": {
      "command": "npx",
      "args": ["-y", "@dplooy/mcp-server"],
      "env": { "DPLOOY_API_KEY": "dpk_live_xxx" }
    }
  }
}

Get an API key from Settings → API Keys. The local server adds one thing remote can't: deploying a folder that already exists on your computer.

The part that matters later

Your site doesn't die when the chat ends

Most AI builders can only create. Ask for a change three weeks later and the assistant has forgotten everything, so it rebuilds — and quietly loses whatever it failed to remember.

Dplooy's tools read your actual live files first, then change only the exact text you asked about.

  • Works in any conversation, on any day
  • Nothing is deleted unless you ask by name
  • Forms, bookings and content survive every edit
  • A stale copy can never overwrite your live site
claude.ai · new chat

Three weeks later

Change our Saturday hours to 9–4 and fix the typo on the about page
get_project_filesreads what's live
update_project_files2 exact edits

Done — live in seconds. Nothing else on the site changed.

Every tool your AI can use

The complete surface your AI can reach. You never type these — your assistant picks the right one. Local installs swap deploy_files for deploy_folder, which reads a folder from your disk.

Hosting

Publish and manage sites

5
deploy_website

Publish a single-page site to a live URL

deploy_files

Publish a complete multi-page site

list_projects

List every site on the account

get_project

Details, URL and stats for one site

delete_project

Permanently remove a site

Editing

Change a live site without rebuilding it

2
get_project_files

Read the live source of a deployed site

update_project_files

Change only the exact text you name

Forms

A real form backend

2
create_form

Create an endpoint + the embed snippet

list_forms

List endpoints and their usage

Content

Editable data without a CMS

4
create_data_collection

Create content from 20 presets

get_data_collection

Fields, rows, URL and fetch snippet

list_data_collections

List collections and where they're used

update_data_rows

Edit single cells, append or delete rows

Bookings

Appointments and reservations

2
create_booking_page

Create a calendar + widget embed

list_booking_pages

List booking widgets and pending requests

Media, chatbot & setup

Images, AI chat and wiring it together

6
upload_media

Add up to 16 images at once, by URL

list_media

List images already in the library

assign_to_project

Connect forms, data and bookings to a site

configure_chatbot

Enable an AI chatbot with a knowledge base

get_account

Plan, usage and available features

get_guide

Full feature rules, read by the AI itself

Built on the current MCP standard

Not a bespoke integration that works with one app today and breaks tomorrow. Dplooy implements the Model Context Protocol authorization spec properly — which is exactly why it works with clients that don't exist yet.

Stateless Streamable HTTP

The current transport, with no handshake to get stuck in. Every request is self-describing, so connections survive restarts, scale cleanly and never end up in a half-open state.

OAuth 2.1 with PKCE

Authorization code flow with S256 challenges — required, never optional. Tokens are opaque, hashed at rest, scoped to your account, and revocable in one click.

Full discovery chain

Protected-resource and authorization-server metadata, resource indicators, issuer identification and dynamic client registration. Your client finds everything it needs by itself.

MCP auth spec 2026-07-28OAuth 2.1 + PKCE (S256)RFC 9728 · protected resourceRFC 8414 · server metadataRFC 8707 · resource indicatorsRFC 9207 · issuer identificationClient ID Metadata DocumentsRFC 7591 DCR fallback

Spec compliance is the compatibility strategy. Any client that speaks remote MCP correctly can connect — today's and the ones released next year.

Refresh your connector now and then

Your AI app loads the tool list once, when you connect — and caches it. We ship improvements and new tools regularly, and a cached list means your assistant simply won't know the new ones exist.

How to refresh

Settings → Connectors → toggle Dplooy off and on (or disconnect and reconnect), then start a new conversation. Takes about ten seconds.

When to do it

Every month or so is plenty — and any time your assistant says it can't find a Dplooy tool it should have. That message almost always means a stale list, not a broken connection.

What you get on each plan

Your AI reads this itself before it designs anything, so it builds within your plan instead of failing halfway through.

Free$0
Websites
3 (expire after 3 days)
Content collections
10 × 25 rows
Media images
10
Form endpoints
Booking widgets
AI chatbot
Custom domain
Plus$4.99/mo
Websites
25
Content collections
200 × 200 rows
Media images
100
Form endpoints
10 · 500 submissions/mo
Booking widgets
25 · 300 requests/mo
AI chatbot
2,000 replies/mo
Custom domain
Pro$12.99/mo
Websites
Unlimited
Content collections
1,000 × 1,000 rows
Media images
500
Form endpoints
50 · 5,000 submissions/mo
Booking widgets
100 · 3,000 requests/mo
AI chatbot
15,000 replies/mo
Custom domain
Included

What it can't do

Worth knowing before you start, so nothing surprises you halfway through a build.

No server-side code

Dplooy hosts static sites. PHP, Python, Ruby and Node backends won't run. The four attachable features cover what most business sites need a server for.

No build step

React, Next.js, Vue and Astro projects must be built first — deploy the output folder, not the source.

AI can't delete your data

It can delete a whole website when you ask it by name — but never a form, content collection, booking widget or image. Removing those is yours to do, from the dashboard.

Images live in the media library

Your AI uploads images by URL into your library and references those. It can't hot-link a third-party image into your site.

Form endpoints are permanent

Connecting a form to a site is one-time and can't be undone — it protects the submissions history. Your AI is told to do it last, after everything else worked.

Deploy limits apply

Publishing is capped per account per hour and per day to stop runaway automation. Normal building and editing never comes close.

No undo

There's no rollback to a previous version. Targeted edits keep changes small on purpose, so mistakes stay small too.

It asks before it assumes

Notification emails, business details and prices come from you. Anything you don't supply, it has to invent — so give it the real details up front.

You never need the AI to change something

Everything your assistant creates is yours, in your dashboard, editable by hand. Here is where each thing lives.

Your websites

Dashboard

Every site your AI published, with its URL, visitor analytics, password protection, custom domain and the in-browser code editor.

Form submissions

Dashboard → Forms

The inbox for every enquiry, with search, spam filtering, CSV export and the notification email address you can change any time.

Editable content

Dashboard → Data

Prices, menu items, team members, FAQs — edit a cell and the live site updates within about a minute. No AI and no redeploy needed.

Booking requests

Dashboard → Bookings

Approve or decline each request, set your availability and hours, and see the calendar. Guests are emailed automatically either way.

Images

Dashboard → Media

Everything your AI uploaded, plus anything you add yourself. Copy a URL to reuse an image on another site.

Connected AI apps

Settings → Connections

Every assistant with access to your account. Revoke any of them in one click — it stops working on its very next request.

Content collections are the one to remember: put your prices, menu or team in one and you can update the live site yourself forever, without opening a chat. How editable content collections work

Letting an AI publish, safely

Giving an assistant access to your hosting should not mean giving up control.

Same scanning as a manual upload

Every AI deploy passes through the identical malware and phishing checks as an upload from your dashboard. There is no separate, looser path.

Revoke in one click

Settings → Connections lists every connected AI app. Revoking cuts it off on its very next request.

Scoped to your account only

A connection can only touch your own projects. It can never see or reach another customer's sites.

Deletion always asks you

Removing a project is an explicit, named action. Nothing is deleted as a side effect of an edit.

Frequently asked questions

What is an MCP server, in plain English?

MCP (Model Context Protocol) is an open standard that lets AI assistants use real tools instead of only writing text. An MCP server publishes a list of actions the AI can take. The Dplooy MCP server gives your AI the ability to build, publish and edit actual websites on live URLs — not code samples you have to deploy yourself.

Do I need an API key for claude.ai or ChatGPT?

No. The remote server uses OAuth: you paste one URL, sign in to Dplooy, and approve the connection on a Dplooy consent screen. No key is ever copied or pasted, and you can revoke the connection at any time from Settings → Connections. API keys exist for local tools and direct REST access, and live in Settings → API Keys.

Is it free?

Yes. The MCP server itself is free and works on the free plan — 3 hosted projects, 10 content collections, 10 media images. Forms, the booking widget and the AI chatbot are Plus ($4.99/mo) or Pro ($12.99/mo) features. Your AI reads your plan before it designs anything, so it builds within what you actually have and tells you what an upgrade would add instead of failing halfway.

Can the AI edit a site it built weeks ago, in a new chat?

Yes, and this is the part most AI website tools get wrong. Any connected AI can read the current live files of any project and change only what needs changing. It does not rely on remembering the site, and it does not rebuild from scratch. Ask for "update my opening hours" in a brand-new conversation and it works.

What can it actually build?

Static websites of any size — landing pages, multi-page business sites, portfolios, menus, documentation — plus four backend features that normally need a server: form endpoints with a submissions inbox and email notifications, editable content collections served as public JSON, a booking calendar with owner approval, and an AI chatbot injected into every page. What it cannot do is run server-side code such as PHP, Python or Node, or run a build step for React or Next.js — deploy the built output instead.

Which AI apps work with it?

claude.ai and ChatGPT connect remotely with no install. Claude Code, Claude Desktop, Cursor, Windsurf and any other MCP-capable client can connect remotely too, or run the npm package locally with npx. On claude.ai, custom connectors require a paid Claude plan, and Team or Enterprise workspaces may need an admin to enable them. In ChatGPT, the full tool set is available through connectors in developer mode.

Is it safe to let an AI publish to my account?

Every deploy an AI makes goes through exactly the same security scanning as a manual upload from the dashboard — there is no separate path. Every action is tied to your connection and logged, you can revoke access with one click, and per-account deploy limits stop runaway automation. Deleting a project always requires the AI to ask you by name.

What is the difference between the remote and local server?

They expose the same tool set and behave identically. The remote server runs on Dplooy's infrastructure, needs no install, and is the only option for claude.ai and ChatGPT on the web. The local server runs on your own machine via npx, uses an API key, and can additionally read files from your disk — so it can deploy a folder you already have locally.

Connect your AI in about two minutes

Create a free account, paste one URL into Claude or ChatGPT, and ask for a website. No credit card, no install, no API key.