Skip to content
BuildBase
Agent-ready apps

Let your users connect Claude Code to your app

Agent-ready apps: make the app you build with the SDK ready for Claude Code. Your users add your app's MCP server to Claude Code, sign in on your hosted page and approve it, and Claude Code acts as them, never as more.

How your users add your app to Claude Code

Give your users your app's MCP address. npx buildbase agent install --url <your MCP address> prints the setup for each AI tool, Claude Code included, with your address filled in.

  1. In Claude Code, your user adds a remote MCP server: claude mcp add. They use your app's MCP address, https://app.example.com/api/mcp in these examples, or the command or config below.
  2. Sign in: run /mcp. Run /mcp in a session and choose the server; a browser opens for sign-in. If it does not open, copy the address it prints. claude mcp login <name> does the same from the shell.
  3. Your hosted sign-in page opens. Your user signs in and approves the consent screen, which names Claude Code and what it asks for.
  4. Claude Code can now use your tools, as that user and with that user's permissions.
terminalBash
claude mcp add --transport http acme https://app.example.com/api/mcp
.mcp.jsonJSON
{
  "mcpServers": {
    "acme": {
      "type": "http",
      "url": "https://app.example.com/api/mcp"
    }
  }
}

The plan it needs

Claude Code does not name a plan for MCP servers. An organization can allow or deny servers with a managed MCP file.

How sign-in looks in Claude Code

Run /mcp in a session and choose the server; a browser opens for sign-in. If it does not open, copy the address it prints. claude mcp login <name> does the same from the shell.

What is different in Claude Code

  • Claude Code runs on your user's own computer, so it reaches any address that computer can, including a development server while you test.
  • Your server's built-in tools are read-only until you allow changes. Claude Code asks before a tool call unless an allow rule says otherwise, and /mcp shows which setting applies to each tool.
  • --scope project writes the server to .mcp.json in the repository, so everyone on the project gets it.

If it does not work in Claude Code

Claude Code cannot register or sign in
Turn on Let agents register themselves, with a base client, in the console under Authentication, then Agent readiness. Claude Code registers its own client the first time it connects.
Claude Code does not connect
claude mcp list shows Connected, Needs authentication or Failed to connect, and claude mcp get <name> adds the reason. MCP_TIMEOUT raises the start-up timeout.
Claude Code needs to sign in again
Run /mcp, choose the server, then Re-authenticate. Clear authentication signs it out. Log out before claude mcp remove, which deletes the local token without revoking it.
Check your server the way Claude Code will
Run npx buildbase agent doctor --url <your MCP address>. It walks the first contact a client makes (the sign-in challenge, the metadata, redirects, a stable tool list) and exits 1 when a check fails.

Frequently Asked Questions

Which Claude Code plan do your users need?

Claude Code does not name a plan for MCP servers. An organization can allow or deny servers with a managed MCP file.

Does my MCP server need to be public for Claude Code?

No. Claude Code runs on your user's computer, so it reaches what that computer reaches. Your users still need an address they can open, and production should be public HTTPS.

Can Claude Code change data in my app?

Only through tools you allow, and only as the user who connected it. The built-in tools are read-only by default, and a tool that needs extra scopes is hidden until the user grants them.

Where next