Client overview

Stackyapper supports OAuth for interactive AI clients and bearer API keys for headless clients that cannot complete an interactive sign-in.

New to connected AI? Start with What is MCP, and why would your business need it? before choosing a client.

The portal shows the same URLs under AI clients → Connect, in the Connect AI software panel, with a Copy URL button for ChatGPT, Claude, Codex, and other HTTP clients, so you can copy the one your client needs rather than typing it.

After connecting, most AI clients list only a small set of stackyapper_ tools — router tools such as stackyapper_search_tools, stackyapper_get_tool_schemas, and stackyapper_execute_read_tool, plus a few workflow and feedback tools. This is the expected surface, not a missing grant: the client uses the router tools to search the full catalog of tools your account is allowed to use and runs them on demand, and Stackyapper checks access again before every request.

ChatGPT also receives the router tools, plus search and fetch tools and a small curated set of common tools with simplified schemas, because ChatGPT rejects large tool lists.

Tool names are pinned for the life of a client session. After an administrator changes your app or tool grants, start a new conversation or reconnect the client so it picks up the new catalog.

Ask the client to run a read-only tool against an app you can read, limited to a few records, then confirm the call appears under Audit in the same workspace with your user and the client name. The full verification flow, including what to do when the test fails, is in Connect a client and verify.

OAuth is preferred because the authorization is tied to the signed-in user and workspace. For automation that cannot open a browser, the signed-in user can create and revoke their own direct API key in the portal. A key inherits that user’s workspace access.

If your workspace requires approval, the connection screen lets you explain what you need the client for and submit an access request. Track it under AI clients in the portal. Reconnecting while a request is pending keeps the original reason and does not send another request. Once an admin approves access, return to your AI client and reconnect. A client that is explicitly blocked by policy needs an administrator to review that policy first.

For a managed team rollout, restrict members to organization-approved MCP servers in your AI provider as well as configuring access in Stackyapper.