Bunstation MCP server

Connect your Bunstation account to Cursor, Claude, Grok Bot, or another compatible AI client. The client can then browse Bunstation templates, start a project from one, create projects from prompts, inspect and edit files, check builds, publish sites, and read account usage without making you copy code between tools.

The remote MCP endpoint is:

https://api.bunstation.com/mcp

Bunstation uses Streamable HTTP and OAuth. Do not add an API key or authorization header: your client opens the Bunstation sign-in and consent flow when it connects.

What MCP means

Model Context Protocol (MCP) is an open standard for connecting AI applications to external tools and data. Bunstation exposes a defined set of tools, so the assistant can discover what it is allowed to do and send structured requests instead of trying to operate the website.

The model and the MCP client are separate. You can use a Grok model inside a compatible client, for example, or connect from the Grok Bot, Claude, or Cursor applications themselves. The client must support remote Streamable HTTP servers and OAuth.

What your assistant can do

List public templates from the Bunstation app library
Filter templates by category and start a project from a selected template
Create a Vite or Next.js project from a prompt
List projects and inspect project details
List, read, and search project files
Stage file creates, replacements, and deletions
Review or discard pending changes before committing
Check builds and read build logs
Publish the latest successful build
Rename a project or change its model
Stop a running generation
Check credits, token usage, plan details, and invoices

Creating projects consumes credits. Bunstation allows the MCP client to create no more than three projects at a time and applies the normal account limits.

Connect Cursor

From Cursor settings

Open Cursor Settings → Tools & MCP (the label may appear as MCP Tools in some versions).
Add a new remote MCP server named Bunstation.
Enter https://api.bunstation.com/mcp as the server URL.
Click Connect or Authenticate, then complete the Bunstation sign-in in your browser.
Return to Cursor and ask: List my Bunstation projects.

With mcp.json

Create .cursor/mcp.json in a project to enable Bunstation only in that workspace, or ~/.cursor/mcp.json to make it available globally:

{
  "mcpServers": {
    "bunstation": {
      "url": "https://api.bunstation.com/mcp"
    }
  }
}

Save the file, open Cursor's MCP settings, and complete the OAuth connection. Keep personal access and tokens out of a project-level file; Bunstation does not require either one in this configuration.

Connect Claude

Claude, Claude Desktop, or mobile

Open Claude Connectors, or go to Customize → Connectors.
Click +, then Add custom connector.
Name it Bunstation and enter https://api.bunstation.com/mcp.
Click Add, then sign in to Bunstation when Claude asks you to connect.
Enable the Bunstation connector for the conversation from the + menu beside the prompt.

You do not need to enter an OAuth Client ID or Client Secret in Advanced settings. The same remote connector is available across supported Claude surfaces after you connect it.

Claude Code

Run:

claude mcp add --transport http bunstation https://api.bunstation.com/mcp

Then open Claude Code and run /mcp. Select bunstation and complete authentication if it is not already connected. Use claude mcp list to check the configured servers.

Connect Grok Bot

Grok Bot on the web

Open grok.com/connectors.
Click New Connector, then choose Custom.
Name it Bunstation and enter https://api.bunstation.com/mcp.
Add the connector and complete the Bunstation OAuth sign-in.
Enable the connector in a conversation and ask Grok Bot to list your projects.

Grok Bot CLI

Run:

grok mcp add --transport http bunstation https://api.bunstation.com/mcp

Grok Bot opens the browser OAuth flow when the server is first used. Run /mcps in Grok Bot CLI to view the connection, authenticate it, enable or disable it, and refresh its tools.

Connect another MCP client

Use these settings in any client that supports remote MCP:

Setting	Value
Name	`bunstation`
Transport	Streamable HTTP
URL	`https://api.bunstation.com/mcp`
Authentication	OAuth
Custom headers	None

If a client only supports local stdio servers or asks you to paste a permanent bearer token, it cannot connect directly to Bunstation. Use a client with remote Streamable HTTP and OAuth support.

Test the connection

Start with a read-only request:

List my Bunstation projects and tell me which one was updated most recently.

Other useful prompts:

Show me the available restaurant templates, recommend one, and wait for me to choose before using it.
Use the pizza restaurant template to create a new project, then show me its files.
Show my Bunstation credit balance and usage for this month.
Inspect the files in my portfolio project and explain how the homepage is structured.
Update the pricing section, show me the pending changes, and wait for my approval before committing.
Check the latest build. If it succeeded, publish it and give me the live URL.
Create a Next.js website for a dental clinic. Use a calm visual style and include online appointment requests.

Give the assistant the project name and a clear outcome. For edits, explicitly ask it to stop after listing pending changes when you want to review them first.

Use Bunstation templates

Ask your assistant to list available templates before choosing one:

List the Bunstation templates for restaurants and summarize the best options.

The templates.list tool returns the same public projects shown in the Bunstation template library, including each template's ID, name, stack, category, thumbnail, and preview URL. The optional category filter accepts values such as restaurant and landing_page.

After you choose a template, the assistant can call templates.use with its ID. Bunstation copies the template repository into a new project owned by your account, exactly like the Use template action in the app. This does not start an AI generation from a prompt; the copied project is ready for inspection and further edits.

Review changes before publishing

File operations use a staged workflow:

The assistant creates, replaces, or deletes files as pending changes.
It lists the pending paths for review.
You can ask it to discard an unwanted change.
It commits the accepted changes, which starts a build.
It checks the build status and logs.
Publishing remains a separate action and only accepts a successful build.

Your MCP client may also ask for confirmation before a tool call. Read the action and its arguments before approving writes, commits, project creation, or publishing.

Troubleshooting

The server appears offline

Confirm that the URL is exactly https://api.bunstation.com/mcp. Do not use the Bunstation app URL, add /sse, or configure a local command.

Authentication does not open

Open the client's MCP or connector settings and choose Authenticate, Connect, or Sign in for Bunstation. Allow browser pop-ups, finish signing in, then return to the client and refresh the server.

The client asks for an API key

No Bunstation API key is needed. Remove custom headers and use the client's OAuth connection action. A client that cannot perform OAuth for a remote MCP server is not currently compatible.

Tools are connected but unavailable in chat

Enable the Bunstation server or connector for that conversation. Some clients configure a server globally but keep its tools disabled per chat or workspace.

Template tools do not appear

Refresh or reconnect the Bunstation MCP server so the client discovers its current tools. Then ask the assistant to list templates. The template tools are named templates.list and templates.use.

A change is not visible on the live site

Ask the assistant to list pending changes, commit them, wait for a successful build, and publish that build. Editing, committing, building, and publishing are separate steps.

Revoke access

Disconnect or remove Bunstation from the client's MCP or connector settings when you no longer want that client to access your account. If you use a shared computer, also sign out of both Bunstation and the AI client.

Editor: work with the same project in Bunstation
Versions: understand commits and builds
Publish: put a successful build online
Plans and credits: understand generation usage