VibeDeploy MCP server

Your AI agent ships the site itself

VibeDeploy runs a remote MCP server at https://mcp.vibedeploy.be/mcp. Add it to Claude, ChatGPT, Cursor, VS Code or any MCP client, sign in once with OAuth, and your agent publishes, edits and redeploys websites and container apps on hosting in Belgium. New accounts get 14 days free without a card; plans start at EUR 4 a month.

MCP server URL
https://mcp.vibedeploy.be/mcp

Streamable HTTP, sign-in with OAuth 2.1. Nothing to install.

OAuth sign-in, no API keys to paste. Hosted in Belgium. 14-day trial, no card.

Create your accountOr connect first: the sign-in window can create the account for you.

Updated

What your agent can do

VibeDeploy gives your agent 40+ tools. These six groups cover most of what you will ask for.

Publish

deploy_site puts a new site online from the files your agent has, and update_site patches one that is already live. When the source is stored on VibeDeploy, build_and_deploy rebuilds and publishes it. Big sites go up in steps: begin_deploy, then add_files or add_file_chunk, then commit_deploy.

  • deploy_site
  • update_site
  • build_and_deploy
  • begin_deploy
  • add_files
  • add_file_chunk
  • commit_deploy
Try

“Deploy the files in this folder as a new VibeDeploy site called launch-page and send me the link.”

Edit live sites

read_files and search_files show what is live. apply_edits and update_file_content make exact find-and-replace edits in one or many files, and edits to the served files go live at once. When the source is stored, the agent edits the source and runs build_and_deploy instead.

  • read_files
  • search_files
  • apply_edits
  • update_file_content
  • read_source_files
  • write_source_files
Try

“Search my-portfolio for the old phone number and replace it on every page.”

Domains

add_custom_domain connects a domain you own and returns the DNS records to set at your registrar. verify_custom_domain checks them, and the SSL certificate is issued automatically.

  • add_custom_domain
  • verify_custom_domain
  • list_custom_domains
Try

“Add www.example.com to my-portfolio and check whether the DNS records are right.”

Forms

set_forms_config switches on the forms relay: an ordinary HTML form on a static site emails every submission to you, without a separate form service. A custom sender address works once its domain is verified.

  • set_forms_config
  • get_forms_config
  • verify_forms_sender_domain
Try

“Send the contact form submissions of my-portfolio to my email address.”

Backends

deploy_app runs any public container image next to the site: your own server from the source tree, Postgres, MySQL, Redis or workers, in a gVisor sandbox. Per site: up to 6 services, 2 CPU and 2 GiB of memory in total, 4 volumes and 3 GiB of storage. get_app_status returns status and logs (the dashboard has no log view yet), and API keys go into the dashboard under the site's Settings > Secrets, never into the chat. Container data is not yet included in snapshots.

  • deploy_app
  • get_app_status
  • remove_app
Try

“Start a Postgres database next to my-portfolio and tell me how my server should connect to it.”

History and analytics

Every deploy is kept as a version, with a restore point taken before each one. list_history shows recent deploys and snapshots, create_snapshot adds a restore point on demand and list_snapshots lists them. Restoring an earlier version is one click in the dashboard's History tab, within your plan's restore window; there is no restore tool over MCP. get_site_analytics returns page views, visitors and top pages from the built-in, privacy-friendly analytics.

  • list_history
  • create_snapshot
  • list_snapshots
  • get_site_analytics
Try

“Take a snapshot of my-portfolio, then show me its last five deploys.”

Install in your client

Pick your client. Every route ends the same way: sign in to VibeDeploy once, then ask the agent to deploy.

Claude: web, Desktop and mobile

One custom connector covers claude.ai, the Desktop app, the mobile apps and Cowork.

  1. Click Add to Claude, or open Customize > Connectors, click + Add and choose Add custom connector.
  2. Name it VibeDeploy and enter the URL below (the Add to Claude button fills in both), then click Continue.
  3. Keep the detected settings (if Claude asks for an OAuth client, choose Register automatically), click Add, and sign in to VibeDeploy when Claude asks.
Remote MCP server URL
https://mcp.vibedeploy.be/mcp

Claude's help center says the Free plan includes one custom connector. On Team and Enterprise plans an owner adds the connector for the organization, and each member then connects it with their own VibeDeploy sign-in.

Guide: put a Claude artifact online

Claude Code

Add the server once for your user account, and every project can use it.

  1. Run the command below in a terminal.
  2. Start Claude Code, run /mcp, pick vibedeploy and sign in in the browser.
  3. To share the setup with a team, commit the .mcp.json file below to the project instead.
Terminal
claude mcp add --transport http --scope user vibedeploy https://mcp.vibedeploy.be/mcp
.mcp.json (project)
{
  "mcpServers": {
    "vibedeploy": {
      "type": "http",
      "url": "https://mcp.vibedeploy.be/mcp"
    }
  }
}

Guide: deploy from Claude Code

ChatGPT

Two routes, depending on what your ChatGPT plan offers.

  1. If your ChatGPT plan lets you add custom MCP servers, open Plugins in ChatGPT on the web, click + and choose Add custom MCP server. Some plans and workspaces need developer mode switched on first; OpenAI's help center describes that switch under Settings > Apps > Advanced settings.
  2. Name it VibeDeploy, enter the URL below, choose OAuth, accept the risk warning and click Create as a plugin. Install the plugin, then sign in to VibeDeploy when asked.
  3. Or paste the deploy guide from your VibeDeploy dashboard into any chat. ChatGPT then sends the files itself, or gives you a deploy page to open in your browser.
Remote MCP server URL
https://mcp.vibedeploy.be/mcp

Publishing needs write access. OpenAI's help center lists full MCP support, write actions included, for Business, Enterprise and Edu workspaces, and says Pro accounts can connect MCP servers with read and fetch permissions in developer mode. Menus differ by plan and workspace, so follow what your ChatGPT shows.

Guide: host a site made with ChatGPT

OpenAI Codex CLI

Codex adds remote servers from the command line or from its config file.

  1. Add the server and sign in with the two commands below.
  2. Or add the block to ~/.codex/config.toml. The longer tool timeout gives large uploads and builds time to finish.
Terminal
codex mcp add vibedeploy --url https://mcp.vibedeploy.be/mcp
codex mcp login vibedeploy
~/.codex/config.toml
[mcp_servers.vibedeploy]
url = "https://mcp.vibedeploy.be/mcp"
tool_timeout_sec = 600

Cursor

One click with the install link, or one entry in mcp.json.

  1. Click Add to Cursor and confirm the install when Cursor asks.
  2. Or add the entry below to ~/.cursor/mcp.json for all projects, or to .cursor/mcp.json for one project.
  3. Sign in to VibeDeploy when Cursor asks, then ask Agent to deploy.
mcp.json
{
  "mcpServers": {
    "vibedeploy": {
      "url": "https://mcp.vibedeploy.be/mcp"
    }
  }
}

Guide: deploy from Cursor

VS Code with GitHub Copilot

VS Code installs the server from a link or from a file in your workspace.

  1. Click Install in VS Code and confirm in VS Code.
  2. Or add the entry below to .vscode/mcp.json in your project.
  3. Trust and start the server, sign in to VibeDeploy when VS Code asks, then ask Copilot in chat to deploy.
.vscode/mcp.json
{
  "servers": {
    "vibedeploy": {
      "type": "http",
      "url": "https://mcp.vibedeploy.be/mcp"
    }
  }
}

Guide: deploy from GitHub Copilot

Windsurf (now Devin Desktop)

Windsurf was renamed Devin Desktop on 2 June 2026, and new chats run on the Devin Local agent. Its MCP servers live in the Devin config file.

  1. Add the first entry below to ~/.config/devin/mcp_config.json on macOS or Linux, or to %APPDATA%\devin\mcp_config.json on Windows.
  2. Or let the Devin CLI write that entry with the command below.
  3. On the first VibeDeploy tool call you are asked to sign in in the browser. If VibeDeploy later shows Needs auth in Devin Local's MCP list, click Authenticate next to it; with the Devin CLI, devin mcp logout vibedeploy followed by devin mcp login vibedeploy does the same.
  4. Windsurf builds from before the rename run the Cascade agent, which reads ~/.codeium/windsurf/mcp_config.json and expects serverUrl instead of url.
Devin Desktop: mcp_config.json
{
  "mcpServers": {
    "vibedeploy": {
      "url": "https://mcp.vibedeploy.be/mcp"
    }
  }
}
Devin CLI
devin mcp add -s user vibedeploy https://mcp.vibedeploy.be/mcp
Older Windsurf builds: ~/.codeium/windsurf/mcp_config.json
{
  "mcpServers": {
    "vibedeploy": {
      "serverUrl": "https://mcp.vibedeploy.be/mcp"
    }
  }
}

Guide: deploy from Windsurf

Gemini CLI

Gemini CLI keeps remote servers in ~/.gemini/settings.json under the key httpUrl.

  1. Add the server for your user account with the command below.
  2. Sign in from a Gemini CLI session with /mcp auth vibedeploy.
Terminal
gemini mcp add --transport http --scope user vibedeploy https://mcp.vibedeploy.be/mcp

Google moved unpaid and Google One users from Gemini CLI to Antigravity CLI on 18 June 2026. We have not tested Antigravity yet, so it is not covered here.

Any other MCP client

Every client that supports remote MCP servers with OAuth can connect.

  1. Point the client at the Streamable HTTP endpoint. It discovers the OAuth server, registers itself and asks you to sign in.
  2. Most clients use a url key. Windsurf builds from before Devin Desktop use serverUrl, Gemini CLI uses httpUrl, and VS Code wraps servers in a servers object with type http.
  3. Clients that cannot sign in with OAuth can send a deploy token from your VibeDeploy dashboard as an Authorization: Bearer header.
Generic config
{
  "mcpServers": {
    "vibedeploy": {
      "url": "https://mcp.vibedeploy.be/mcp"
    }
  }
}

How it works in 3 steps

  1. 1

    Add the URL

    Use a button above or the steps for your client. The address is always https://mcp.vibedeploy.be/mcp.

  2. 2

    Sign in once

    A VibeDeploy sign-in window opens. Google, GitHub or Microsoft is fastest, and the same window creates an account if you do not have one yet.

  3. 3

    Say "Deploy this site to VibeDeploy"

    The agent uploads the files and answers with the live address. Later chats can edit and redeploy the same site.

Prompts to start with

Copy one into your AI client after connecting. Replace my-portfolio with your site name.

Security and data

OAuth 2.1 sign-in

PKCE (S256) and dynamic client registration. Your AI client gets its own token, so you never paste a password or API key into a chat.

Short-lived tokens

Access tokens expire after an hour and refresh tokens rotate on every use. A token only reaches the sites of your own team.

Guard rails on destructive tools

Tools that replace or delete live files are annotated as destructive. delete_site, and deploy_site on an existing site, need an explicit confirmation; update_site and build_and_deploy can replace every file in one call without one, so review their calls before you approve them. A restore point is taken before every deploy.

You decide who keeps access

Remove the connector in your AI client to end its access. Each client signs in separately, so removing one leaves the others untouched.

Rate limit

200 tool calls per hour per token.

Hosted in Belgium

Sites, files and container apps run on VibeDeploy's own cluster in Belgium, operated by Tnet Comm.V.

The subprocessors page lists all 8 subprocessors and what each one handles. The privacy policy explains what your AI assistant's provider receives when you use the tools.

Limits and pricing

Every plan gets the same MCP server and tools. The plan sets how much you can host.

VibeDeploy plans and limits
PlanTrialSoloMakerStudio
Price per month14 days, no cardEUR 4EUR 15EUR 39
Price per yearn/aEUR 38EUR 144EUR 374
Sites11310
Custom domains1 (link one you own)1310
Storage500 MB1 GB2 GB5 GB
Bandwidth per month1 GB5 GB10 GB50 GB
Restore window (one click in the dashboard)3 days3 days3 days7 days
StagingNoNoNoYes
Private and password-protected sitesNoNoYesYes

Prices include VAT where it applies. Yearly prices are 20 percent below twelve monthly payments, rounded down.

The trial lasts 14 days and needs no card. You can link a domain you already own; buying a domain through VibeDeploy needs a paid plan. If the trial ends without a plan, the site goes offline and your files are kept for 30 days.

Containers run on every plan, the trial included. Limits per site: up to 6 services, 2 CPU and 2 GiB of memory in total, 4 volumes and 3 GiB of storage.

Need more? Business (50 sites) is on /business. All plans side by side: pricing.

Questions about the VibeDeploy MCP server

Do I need an account before connecting?

No. When your AI client first connects, a VibeDeploy sign-in window opens, and it can create the account on the spot. Google, GitHub or Microsoft sign-in is the quickest. New accounts start a 14-day trial without a card.

Does it work on Claude's Free plan?

Claude's help center says the Free plan includes one custom connector, so VibeDeploy can be that connector. Custom connectors work in Claude on the web, in Desktop, on mobile and in Cowork, although Cowork itself needs a paid Claude plan.

Is VibeDeploy in the Claude or ChatGPT directory?

Not listed yet. Until it is, add it as a custom connector with the button above.

Can my agent run a backend or a database?

Yes. deploy_app runs any public container image next to a site, such as your own server built from the source, Postgres, MySQL, Redis or a worker, in a gVisor sandbox. Each site can have up to 6 services with 2 CPU and 2 GiB of memory in total, 4 volumes and 3 GiB of storage. Containers reach the internet but not private networks or SMTP port 25. Container data is not yet included in snapshots.

How do large sites upload?

The agent opens a staged deploy with begin_deploy, sends files in batches with add_files (gzip and base64 keep them small), splits a single large file with add_file_chunk, and publishes everything at once with commit_deploy. If its sandbox can reach the internet, it can instead upload one gzipped tar archive of up to 100 MB (200 MB once unpacked) to the upload address that begin_deploy returns. A single file can be up to 100 MB and a staged deploy up to 500 MB in total; your plan's storage still applies.

Which transport and spec does the server use?

Streamable HTTP at https://mcp.vibedeploy.be/mcp, MCP protocol version 2025-11-25. Authorization is OAuth 2.1 with PKCE (S256) and dynamic client registration, discovered through protected resource metadata (RFC 9728) and authorization server metadata (RFC 8414). initialize and tools/list answer without a token; tool calls require sign-in.

Is there a local or stdio version?

No. VibeDeploy is a remote server only, so there is nothing to install or update. Scripts and tools without MCP support can deploy through the REST API with a deploy token from your dashboard.

Where is my data hosted?

On VibeDeploy's own Kubernetes cluster in Belgium, operated by Tnet Comm.V. Site files, versions and container data stay there. The AI assistant you connect is run by its own provider, which processes what passes through your conversation under its own terms. All 8 subprocessors are listed on the subprocessors page.

What is the registry name?

be.vibedeploy/vibedeploy in the official MCP Registry. Directories that read the registry, such as Glama, list the server under the same name.

Listed in

Find the server outside this site:

New to the idea? Read what MCP hosting is.

Connect once, then ship from the chat

Add the server to the AI tool you already use and let it publish. 14-day trial, no card; plans from EUR 4 a month.