Cloud/Connectors/Connection Tokens
Connection Tokens
One key per AI client: create a named token for Claude Desktop or ChatGPT, watch the request count each one burns, switch any of them off in a click and revoke the ones you no longer trust.
Here is a rule worth internalizing: one token per AI client. Your Claude Desktop gets its own token, ChatGPT gets another, and every request each one makes is counted under its own name. The Connection Tokens tab is where those keys live. The console introduces it exactly as it is: "Manage API tokens for your AI clients to connect to this connector."
Connection Tokens
Manage API tokens for your AI clients to connect to this connector.
| Client | Token | Requests | Status | Last used | Actions |
|---|---|---|---|---|---|
| vk_live_4f9a•••• | 1,204 | 2m ago | |||
| vk_live_8c2d•••• | 486 | 1h ago | |||
| vk_live_1b7e•••• | 0 | Never |
The mockup is the real editor: click New Token, pick a client chip or type a name, create it; then flip a Status switch and try Revoke with its confirmation. Nothing leaves this page.
Why one per client
When every client has its own key, the table becomes a per client report: the Requests column shows exactly how much each one consumes, Last used shows which ones are alive, and if one client misbehaves you switch that one off without touching the others. Share one token with everything and you are blind: an anomaly anywhere is an anomaly everywhere.
Creating a token: named and suggested
New Token opens a dialog that asks for a Token Label with a smart twist: "Suggestions — or type any name you like" offers one click chips for the usual suspects (Claude Desktop, Claude.ai, ChatGPT, Gemini, Microsoft Copilot, Perplexity, Raycast). Pick a chip and the token is labeled and linked to that client's Quick Connect guide. The dialog shows the connector it belongs to and your quota ("3 of 5 tokens in use" on the sample), and a green note sets expectations: "Scoped to this connector only. You can revoke it at any time."
The table: usage, switches and revocation
Each row is one client key: the client name (with its icon when it is a known one), the token hint masked as vk_live_•••••••• (never the full key), the Requests it burned, a Status switch, Last used (Just now, 2m ago or Never) and Revoke.
- The Status switch turns a client off in one click without deleting anything: the row grays out and wears a Disabled badge, and that client immediately loses access while the others keep working.
- Revoke asks once: "Revoke this token?" with the consequence in plain words: "This action cannot be undone. Any AI client using this token will immediately lose access to this connector." Confirm and the key is gone from the table and from the world.
The kill switch connection
These are the same tokens that appear in Access Tokens across AI Governance, and the same keys your Dashboard attributes activity to. If a link leaks, the playbook from Install Catalog applies: revoke the token here and hand out a new one. The old link stops working instantly; the fleet keeps running.
The empty state is the console's own: "No tokens yet. Create one to connect your AI client." And on the Free plan the button reads Unlock tokens, the console's upgrade path, because minting keys is a paid feature.