Build with an AI agent

Describe an app to your AI assistant. It can create a Buildra project, work through build questions, show you the preview, and publish when you ask.

Buildra’s remote MCP server connects your assistant to your existing Buildra account. You sign in and choose the permissions it receives. Build turns use your credits, with the same spending caps as the build page.

Connect your assistant

Use this server URL with a client that supports remote MCP over Streamable HTTP and OAuth:

https://mcp.buildra.dev/mcp

You need an existing Buildra account. When your client opens the authorization flow, sign in on app.buildra.dev, review the client name and requested permissions, then select Allow or Deny. An agent cannot create an account for you.

Claude Desktop

Open Settings → Connectors in Claude and add a custom connector named Buildra.
Enter https://mcp.buildra.dev/mcp as the remote server URL.
Connect, complete the Buildra sign-in and consent, then enable the connector in your conversation.

Custom connector availability can depend on your plan and workspace settings. See Claude’s custom connector guide.

Claude Code

Run the command, then open /mcp in Claude Code to authenticate:

claude mcp add --transport http buildra https://mcp.buildra.dev/mcp

Complete the browser consent and return to your session. The available tools reflect the permissions you approved.

Cursor

Add the following entry to your project’s .cursor/mcp.json, or merge it into ~/.cursor/mcp.json for all projects. Keep any other servers already configured.

{
  "mcpServers": {
    "buildra": {
      "url": "https://mcp.buildra.dev/mcp"
    }
  }
}

Open Cursor’s MCP settings, enable Buildra, and complete the OAuth connection when prompted. See Cursor’s MCP documentation.

ChatGPT

Enable developer mode in Settings → Apps → Advanced settings, if available. A managed workspace may require an administrator to enable custom MCP apps.
Create a custom app named Buildra, enter https://mcp.buildra.dev/mcp, and choose OAuth authentication.
Connect your Buildra account through the consent flow, then select the app in a conversation.

This is a custom connection; it does not require a Buildra directory listing. Availability and controls depend on your plan and workspace. See OpenAI’s developer mode guide.

From prompt to published app

Try: “Build a habit tracker with Buildra, with a weekly summary. Show me the preview before publishing.”

Create. Your assistant calls create_project, which creates the project and starts its first build turn. For an existing project, it uses send_message.
Wait and answer. It calls wait_for_turn until the turn ends or needs an answer. A running result means wait again. A pending question needs answer_question; the build keeps waiting until it is answered.
Preview. It uses get_preview to get a temporary share link. Web and API projects can have a URL; phone projects may also have an Expo URL. Terminal and none projects have no preview URL.
Publish. When you ask to publish, it calls publish, then checks get_publication for the result. Acceptance of the request does not mean the deployment has finished.

For the complete tool workflow, load the Build with Buildra skill. You can also ask your assistant to read project status, transcripts, and your credit balance, or to stop a running turn.

Permissions you control

Your client can request a subset of these scopes. The consent page shows only those requested; tools outside the approved scopes are unavailable.

Scope	Consent text	What it covers
`projects:read`	See your projects and what the build agent is doing	Project lists, status, transcripts, and preview links.
`projects:write`	Create new projects	Creating a project also needs the build permission to start its first turn.
`build`	Talk to the build agent on your behalf (spends your credits)	Send messages, answer questions, and stop turns.
`publish`	Publish projects to buildra.page	Request a publication and read its status.
`billing:read`	See your credit balance	Read your tier, balance, and project quotas.

OAuth is issued by https://api.buildra.dev. Your client discovers it through the MCP resource metadata. Connections can refresh their access until they expire or you revoke them.

Credits and limits

create_project starts the first build turn and send_message starts or queues another. Both spend your credits. Your balance and per-turn cap still apply; reading status, waiting, and answering a pending question do not start another build turn.

Free: up to 30 build turns per approved agent connection per UTC day.
Pro: up to 300 build turns per approved agent connection per UTC day.
The service’s default rate limit is 60 tool calls per minute per connection.

Daily turn ceilings are not free credit allowances. When your balance runs out, the agent receives out-of-credits. At the daily ceiling it receives agent-daily-limit, with the usage and reset time. You can review credits in Billing and agent usage in Connected agents.

Preview links last 30 minutes by default and at most 60 minutes. Publishing is subject to your account’s project and hosting limits. Projects that cannot be published return not-publishable.

Revoke a connection

Open Settings → Connected agents in Buildra. Review the client name, permissions, connection date, last use, and today’s turns. Select Revoke and confirm.

Subsequent requests and token refreshes are refused. Reconnecting requires your authorization again. Revocation does not delete your projects or unpublish an app.

For help with a connection, email support@buildra.dev.

For agents and client developers

Read the MCP server card, registry document, and skills index. The A2A card is a discovery advertisement only. Buildra serves MCP; it does not accept A2A requests.

This guide is available in English. Request Accept: text/markdown at this URL for its Markdown edition.