Skip to content
Prysm:ID Docs

Connect Claude Code in 60 seconds

Setting up Prysm:ID from Claude Code is literally copy + paste. The dashboard generates an instructional prompt personalized to your workspace, and the agent walks through the steps until the MCP server is registered, authenticated, and your workspace is ready to integrate.

  1. Create a workspace.

    If you don’t have one yet, go to app.prysmid.com and create a workspace (slug + display name + plan). Provisioning takes ~30 seconds. Once it reaches state active, continue.

  2. Copy the setup prompt from the dashboard.

    In the Overview or Workspaces tab you’ll see a “Connect your agent” section. Pick the workspace, language (ES or EN), and click “Copy setup prompt” on the Claude Code card.

    The prompt comes with your display_name, slug, and auth_domain already interpolated — no placeholders to fill in.

  3. Paste it into a fresh Claude Code session.

    Open Claude Code in any folder and paste the full prompt. The agent will:

    • Check Node ≥ 20 and the claude CLI
    • Back up your ~/.claude.json
    • Register the MCP server with claude mcp add --scope user prysmid -- npx -y @prysmid/mcp
    • Complete the device flow (URL + 8-character code in XXXX-XXXX format that you confirm in the browser). If your agent can launch background shell processes, it does this on its own and shows you the URL + code directly in the chat. Otherwise, it’ll ask you to open a terminal and run npx -y @prysmid/mcp once.
    • Ask you to restart Claude Code so it picks up the MCP
    • Verify the tools appear and authentication is OK

  4. The agent continues with the conversational setup.

    Once the MCP is connected, the same prompt continues to guide the agent through:

    • Calling prysmid_setup_check and showing you what’s missing
    • Asking for Google OAuth credentials (if you want Google login) and configuring it with enable_google_login
    • Creating your product’s OIDC app with create_oidc_app
    • Generating the wiring for your framework (Next.js + Auth.js, FastAPI + Authlib, etc.) in your repo, including a .env.local with PRYSMID_ISSUER, PRYSMID_CLIENT_ID, PRYSMID_CLIENT_SECRET, PRYSMID_REDIRECT_URI
    • Re-running prysmid_setup_check and reporting verdict: ready once everything is wired

    Your role during this: paste credentials when asked, and confirm destructive actions. The agent doesn’t make up redirect URIs or app names — it asks.

Other agents (Antigravity, Cursor, etc.)

Section titled “Other agents (Antigravity, Cursor, etc.)”

The dashboard also has a prompt for Google Antigravity (same flow, different registration command). For generic MCP clients (Cursor, Claude Desktop, others), the panel has a manual JSON snippet:

{
  "mcpServers": {
    "prysmid": {
      "command": "npx",
      "args": ["-y", "@prysmid/mcp"]
    }
  }
}

Paste it into your MCP client’s config, restart, and the first tool call triggers the device flow.

Troubleshooting

Section titled “Troubleshooting”

The agent reports 401 Unauthorized on the first tool call. The device flow didn’t complete. Go back to the prompt’s step “open a terminal and run npx -y @prysmid/mcp” — confirm the URL in the browser, wait until the terminal prints device flow login complete.

claude mcp list shows prysmid: ✗ Failed. Check Node ≥ 20 (node --version) and npm connectivity (npm ping). The MCP can’t start without those two.

The Claude Code session hangs without ever asking for device flow. Claude Code launches the MCP as a stdio subprocess, and stderr isn’t a TTY — so the device flow prompt won’t appear unless you pre-cache it in a real terminal. The dashboard prompt includes this step explicitly — follow it in order.

The agent wants to delete something (delete_workspace, delete_oidc_app, delete_idp). Those tools require explicit confirmation from you. If you don’t want it deleted, say “no” — the agent won’t auto-confirm.

Next steps

Section titled “Next steps”