Claude Code

Download the Starter Project

From your project’s Details page in Buildpad.ai, find the AI IDE Setup section and click Download Starter under the Claude Code card.

The download contains a ready-to-use project with agent instructions, agent skills, subagents, and backend connections — everything Claude Code needs to understand your project and start coding.

Explore the Project Structure

The starter project is generated with a root CLAUDE.md, a .claude/ directory, and a project-scoped .mcp.json:

• project-starter
  • CLAUDE.md
  • .mcp.json
  • .claude
    • skills
    • agents

Key components:

• CLAUDE.md — Core rules and guidelines that Claude Code follows when assisting with your project. Claude Code treats this file as project memory  and loads it automatically into every session.
• .claude/skills/ — Skills  are reusable instruction packages that follow the open Agent Skills  standard. Invoke them as slash commands (type / to browse) or let Claude Code activate them automatically when your request matches what a skill does. 57 skills are included, covering the spec-driven development workflow (/buildpad-spec-*), creating API routes, generating tests, scaffolding features, managing access control, and more.
• .claude/agents/ — Subagents  that run focused work in their own isolated context. Claude Code delegates to them automatically by their description — hand a finished slice to code-reviewer, coverage work to test-engineer, or auth/security checks to security-auditor, and run independent tasks in parallel.
• .mcp.json — MCP (Model Context Protocol)  is an open standard that lets AI assistants connect to external tools and data sources. This project-scoped config connects Claude Code to the Buildpad component library, your live DaaS backend, and the Buildpad platform.

Connect to Backend Services

The .mcp.json file at the project root connects Claude Code to three backend services:

• buildpad — Gives the agent access to the Buildpad UI component library, API patterns, and platform conventions so it can write code that follows your project’s standards.
• daas — Connects the agent to your project’s live backend, giving it direct access to your database collections, schemas, and runtime environment for accurate code generation.
• buildpad-platform — Connects the agent to the Buildpad platform, giving it access to your full project context: organization, project configuration, microapps, and environment URLs.

{
  "mcpServers": {
    "buildpad": {
      "command": "npx",
      "args": ["-y", "@buildpad/mcp@latest"]
    },
    "daas": {
      "type": "http",
      "url": "https://<your-project-id>.daas.buildpad.ai/api/mcp",
      "headers": {
        "Authorization": "Bearer <your-daas-token>"
      }
    },
    "buildpad-platform": {
      "type": "http",
      "url": "https://app.buildpad.ai/api/tools/mcp",
      "headers": {
        "Authorization": "Bearer <your-mcp-token>"
      }
    }
  }
}

Open the Project and Approve the Backend Connections

For security, Claude Code asks for approval before using project-scoped servers from .mcp.json. How you open the project depends on which version you use — the desktop app  and the CLI share the same configuration (CLAUDE.md, MCP servers, skills, and settings), so everything below applies to both.

Desktop app:

1. Open Claude and click the Code tab at the top center.

2. Select Local as the environment, click Select folder, and choose the extracted starter project directory.

3. Type /mcp in the prompt box to open the MCP server list. If the buildpad, daas, and buildpad-platform servers appear as pending, approve them there.

4. If the servers don’t show up, ask Claude directly in the prompt:

   Connect the MCP servers defined in .mcp.json

   Claude Code will register the project servers and request your approval for each one — accept them to activate the backend connections.

5. Run /mcp again to confirm all three servers are connected and their tools are available. You can also open the integrated terminal (**Ctrl+**) and run claude mcp list` to check from the command line.

Terminal (CLI) or IDE extension:

1. Open a terminal in the project and run claude (or open the project in your IDE with the Claude Code extension).
2. On first run, Claude Code prompts you to review and approve the three servers — accept them to activate the backend connections.
3. Type /mcp at any time to check connection status, or run claude mcp list from the terminal.

Once connected, the agent can query your database, look up component documentation, and generate code that works with your actual data.

Start Building with AI

With backend connections active, start describing what you want to build — or invoke a skill directly: type / in the prompt box (in the desktop app you can also click + → Slash commands to browse them):

/create-project Design a task management application with teams and projects

After the foundation is up, features follow spec-driven development: /buildpad-spec-init → /buildpad-spec-requirements → /buildpad-spec-design → /buildpad-spec-tasks → /buildpad-impl, with a review gate at each step (start with /buildpad-discovery when scope is unclear).

The agent will use the instructions, skills, and backend connections to:

• Create projects — Bootstrap a full Next.js application with DaaS integration.
• Scaffold features — Generate CRUD pages, forms, and API routes from your collection schemas.
• Generate tests — Create Playwright end-to-end tests for your application.
• Manage data — Create database migrations, collections, and set up role-based access control.
• Add UI components — Install and configure Buildpad UI packages for forms, tables, and more.
• Delegate in parallel — Hand reviews, security audits, and test authoring to the bundled subagents while the main session keeps building.