Skip to main content

Before you begin

You’ll need:
  • A Stardeck account and membership in the org you want to connect to
  • A role to bind the connection to — see Members & Roles. Pick the least permissive role that covers what you need.
  • Access to your org’s Settings → AI Integrations tab

The authorization flow

Every connection goes through a one-time browser authorization:
  1. Your AI client discovers the Stardeck MCP server (you provide the URL)
  2. Stardeck opens a browser page where you sign in, pick an org, pick a role, and choose read or read+write access
  3. Your client receives a token and is ready to use
The server URL is the same for every client:
https://www.stardeck.ai/api/mcp
After authorization you can change the role or access level — or revoke access entirely — from the AI Integrations tab in your org’s Settings. You don’t need to re-authorize to change the role; the change takes effect on the next tool call.

Client-specific setup

Claude Code (the CLI) supports OAuth MCP servers natively.

Add the server

Run this command once in your terminal — you can run it from any directory:
claude mcp add --transport http stardeck https://www.stardeck.ai/api/mcp
This writes an entry to your global ~/.claude/claude_desktop_config.json (or the project-level .mcp.json if you prefer scoping it to a single checkout). Name it whatever you like — stardeck is conventional.

Authorize

Next time you open Claude Code, it will prompt you to approve the new server and then open your browser to sign in. Complete the authorization there.To trigger it immediately:
claude
Then type /mcp to see connected servers and their status. A stardeck entry with a green indicator means it’s live.

Verify

Ask Claude something like:
List the items in my knowledge store
or:
What apps does my org have?
If you get real data back, you’re connected.
The stardeck server here is the org-level connection. If you’re also using Claude Code for local development on a specific app, you may have a separate project-level connection in your app’s .mcp.json — that’s a different server scoped to that one project. Both can be active at the same time.

Managing connections

All active connections appear in Settings → AI Integrations in your org dashboard. From there you can:
  • Change the role the connection acts as — takes effect immediately, no re-authorization needed
  • Change the access level (read only ↔ read & write)
  • Revoke access — the connection stops working on its next request
Members can always manage their own connections. To manage another member’s connections, you need the Manage AI Connectors system permission.

Troubleshooting

Authorization opens but shows an error after sign-in

Make sure you’re signed into the Stardeck organization you want to connect. If you belong to multiple orgs, pick the right one in the org picker.

”This connector has no tools available”

The role you picked doesn’t hold any agent permissions. Go to Settings → AI Integrations, change the role to one with agent:* permissions (e.g. a member or admin role), and the change takes effect immediately.

Tools appear but produce “Access denied” errors

The role has the right agent permission but the per-object check failed — for example, you’re querying a data store the role doesn’t have a grant for, or a knowledge item in a folder the role can’t access. Check the data store’s Access settings or the knowledge folder’s permissions.

Connection disappears after the next restart

Some clients require re-authorization if the server isn’t reachable during startup. Make sure the Stardeck server URL is correct and that the token hasn’t been revoked. Run the claude mcp add command again if needed.

Next steps

Tools & Permissions

Full reference of tool groups and the permissions that unlock them

Members & Roles

Create and configure roles for AI connections

Cross-App Communication

How apps call each other — the same mechanism the AI tools use

Starcat

The in-dashboard agent that uses the same capabilities