Skip to content
All changes

Manage BuildBase from your AI tools

Connect Claude, ChatGPT, Cursor or another AI tool to BuildBase, sign in as yourself, and work in your organization from the chat, within your role.

Added
15
  • AgentsYou can now work in your BuildBase organization from Claude, ChatGPT, Cursor, VS Code and other AI tools. Add https://api.console.buildbase.app/mcp to the tool as a remote MCP server. MCP is the standard way AI tools connect to other services.
  • AgentsThe first time it connects, the tool opens BuildBase's sign-in page and a consent screen, and you approve it. The agent then acts as you, with your role. It never sees API keys, sign-in settings or payment credentials.
  • ConsoleSecurity, in your account menu, has a new AI agents card. Copy the address there, or use Add to Cursor, Add to VS Code, or the ready-made Claude Code and Codex commands. Get started has an optional Connect your AI step that opens it.
  • AgentsThe server offers 19 tools. An agent can read what the console shows you, create and change things, and export your audience. It can also search the BuildBase docs and list the environment variables an app needs (get_env_template).
  • BlueprintsOwners and admins can have an agent write a blueprint, one file that describes an organization's setup. The agent shows what applying it would change before it applies it. Agents can read your plans, add-ons and credit packs but never change them, so a blueprint that would is refused up front.
  • AgentsAn agent can create an organization for you with create_org, once you approve its name. You own it, and it starts on the same plan as one you create in the console.
  • AgentsA CI job or script with nobody to sign in can connect with one of your organization's API keys instead. The key works only in its own organization, and the AI agents setting still applies to it.
  • SecurityEvery agent starts read only, with the console:read scope, a named permission you grant. When a task needs more, BuildBase tells your AI tool which scope is missing. You then approve just that one: console:write, console:export or console:destructive.
  • SecurityRisky changes go through their own tool, console_destructive_change, and need the console:destructive scope. They include deletes, changes to members, roles or webhooks, sends to an audience, moving money and running a workflow for real. An AI tool that can ask you directly shows you each one, and your approval works once, within five minutes.
  • ConsoleOn the same card, owners and admins choose what agents may do in the organization: Off, Read only or Read and change. Read only is the default. A change reaches every connected agent at once.
  • ConsoleSecurity lists the agents connected to the organization, and Block stops one at once. Owners and admins see and block every member's agents. In the audit log, an agent's change is marked AI agent, with the AI tool's name and the member it acted for.
  • WebsiteA new page, buildbase.app/mcp, shows how it works, with a setup page for each AI tool. The menu now tells BuildBase in your AI tools apart from Agent-ready apps, where your own users connect AI tools to your app.
  • DocumentationThe docs have a page for each of 16 AI tools. Each says where to add the server, which plan the tool needs and what signing in looks like there.
  • WebsiteAI agents can find the server by themselves. The website lists it in llms.txt and its API catalog. Both buildbase.app and docs.buildbase.app describe it and its sign-in under /.well-known/.
  • WebsiteThe website and the docs offer read-only WebMCP tools, which a web page gives to an AI agent in the browser. With them an agent searches either site and reads a page as Markdown. The home page also has a Markdown version, buildbase.app/index.md.

Use what shipped this week

Everything in this release is already in the console and the SDK.

Free plan7-day free trial on paid plansNo credit card required