Use BuildBase in Claude Code

This page is part of BuildBase in your AI tools: it connects Claude Code to BuildBase's own MCP server so you can manage your BuildBase organization from the terminal. Claude Code signs in as you and never sees your organization's secrets.

What you need

• Claude Code.
• A BuildBase console account, whichever way you sign in to the console.
• An organization whose AI agents setting is not Off (choose what agents may do). Agents can read by default.

Add the server

claude mcp add --transport http buildbase-console https://api.console.buildbase.app/mcp

Add --scope project to write it to .mcp.json in the repository and share it with everyone on the project:

{
  "mcpServers": {
    "buildbase-console": {
      "type": "http",
      "url": "https://api.console.buildbase.app/mcp"
    }
  }
}

Keep "type": "http": an entry with a url and no type is read as a local (stdio) server and fails.

Sign in

1. Run /mcp inside Claude Code and choose buildbase-console, or run claude mcp login buildbase-console.
2. Your browser opens BuildBase's sign-in page. Sign in with your console account and your usual second step.
3. Read the consent screen: it names the client and the scopes it asks for, console:read and console:write to start. Approve it.

claude mcp list shows Needs authentication for a server still waiting for sign-in. When a task needs more, such as an export or a delete, Claude Code is told which scope is missing (what an agent asks for).

What you see when you connect

Once you have signed in, BuildBase shows the consent screen in your browser. It names the tool, lists the scopes it asks for and the server the access is for, and changes nothing until you choose Allow access. Here it is for a tool that registered itself:

Use it while you build

Ask, for example:

• "Which environment variables does a Next.js app need for Acme? Write them to .env.example, secrets left blank."
• "Write a blueprint for our roles, email templates and forms, and show me the plan."
• "Search the BuildBase docs for magic-link sign-in and add it to this app."

More, with what each one asks you to approve: example prompts.

Without a person: an API key

In CI or a scheduled job there is nobody to sign in. Send one of your organization's API keys as a header instead:

claude mcp add --transport http buildbase-console https://api.console.buildbase.app/mcp \
  --header "Authorization: Bearer $BUILDBASE_API_KEY"

The key works in its own organization only, the AI agents setting still applies, and changes are recorded as the key's (connect with an API key).

What differs in Claude Code

• It runs on your computer, so the sign-in opens your local browser.
• Tool permissions. Claude Code asks before a tool runs unless you allowed it. Allow the read tools if you like; keep console_destructive_change, create_org and blueprint_apply on asking.
• Approvals of a change. Where Claude Code can ask you a question itself, the change BuildBase wants to make is shown to you and nothing happens unless you approve it. Otherwise it states the change and confirms it with a one-time code (what stays with you).
• Large answers. Claude Code warns above 10,000 tokens of tool output and caps it at 25,000 by default (MAX_MCP_OUTPUT_TOKENS). Ask for a narrower read, such as one page of a list, when it cuts one off.

If something goes wrong

To disconnect, run /mcp, choose buildbase-console and Clear authentication first (claude mcp remove deletes the local token without revoking it), then claude mcp remove buildbase-console, and disconnect the agent in BuildBase, which ends its session at once.

Related

• BuildBase in your AI tools: the server, its tools, and what stays with you.
• Use BuildBase in Codex and Use BuildBase in Cursor.
• Let your users connect Claude Code to your app: the same client in Agent-ready apps.
• Anthropic's own guide: MCP in Claude Code.