The agent’s key is its identity — a scoped, revocable credential held on your server. It never reaches the model, the browser, a prompt, or any downstream MCP call. Only a short-lived token does.
Flow
1
Provision an agent
Provision an agent from your dashboard. You receive an
agentId and one dedicated key — the anchor for the agent’s quota, billing, and revocation. Revoke it to cut the agent off. See Credentials & Authentication.2
Authorize with the key
Your server uses the agent’s key to obtain a short-lived token for the connection. WalletSuite’s token provider handles this — the raw key stays on your server, and the token is refreshed before it expires.
3
Connect the MCP client
Attach the token as
Authorization: Bearer <token> on your MCP client, pointed at the hosted endpoint. The agent sees only its bounded tool surface.Connect
Point your MCP client at the hosted endpoint and attach the token. This works for any framework or custom runtime that speaks MCP over HTTP.The token is short-lived and opaque — your client attaches it and refreshes it, and never needs to parse or persist it. The key is the durable secret; the token is disposable.
Managed and embedded assistants
Managed and embedded assistants are authorized by WalletSuite’s own server after an assistant-invoker binding check. Your app code selects the
assistantId — never the model — and the model never sees a key or a token.Next
Agents & Authority
The five bound things that make an agent, and why the model never authorizes.
Execution Levels
What a connected agent can do, from read-only to signing.
Policy Gates
Deny-by-default rules that decide what an agent is allowed to sign.
Choose Your Surface
The five surfaces of the one hosted platform.