Academy
5. Platform-Wide Usage: ChatPro, Studio, Workflows & Beyond
Understand how a single MCP connection surfaces as tool badges in ChatPro, available actions in OUPI Studio and the Assistant, workflow steps, RAG imports, and Deep Research.
A single MCP connection is not scoped to one product — it surfaces everywhere the AI acts on OUPI. When you authorize a service in the MCP Hub, the resulting connection and its tools become available as:
• ChatPro — tool badges appear directly in the chat form, letting the model invoke external actions mid-conversation. • OUPI Studio & the Assistant — agents can call the same tools during autonomous runs. • Workflow steps — any workflow can include an MCP tool call as a discrete step. • RAG imports — pull data from connected services into your knowledge bases. • Deep Research — the research engine can query connected services for live information.
Connect once, use across the entire platform. No per-product configuration is needed.
Every connection exposes a deterministic tool list — the exact set of actions the AI is allowed to perform. Open any connection in the Hub to inspect its tools: each one describes the action it performs, its parameters, and its return shape. This is not a simplified summary; it is literally what the model sees at inference time when deciding whether to invoke the service.
This transparency matters for advanced use cases: if a tool you expect is missing, the granted OAuth scopes may be too narrow. Re-authorize with broader scopes, and the tool list updates accordingly.
Every tool invocation — regardless of which OUPI surface triggered it — is recorded in the execution log. Each entry captures:
• Tool — which action was called. • Time — exact timestamp. • Origin — whether the call came from ChatPro, Studio, a workflow, RAG, or Deep Research. • Outcome — success or failure, with details.
This log is your single audit trail. Use it to debug failing integrations, verify that agents are calling the right tools, and monitor usage patterns across surfaces.
Tool calls do not carry a separate cost line. Their credit consumption is folded into the AI operation that triggers them: a ChatPro conversation, an agent run in Studio, or a workflow execution. This means you do not need to budget MCP usage independently — just be aware that complex tool chains within a single operation will increase that operation's total credit spend.
If a connection breaks, the cause is almost always an expired or revoked OAuth token on the service side. Re-authorize or recreate the connection in the Hub. Deleting a connection removes it platform-wide, so any workflow or agent referencing it will lose access immediately.
Before wiring an MCP connection into a production workflow, open the connection in the Hub and review its tool list. Confirm the scopes match what your workflow needs — missing scopes are the #1 reason a tool call silently fails. Also check the execution log after the first run to verify outcomes.
When you delete a connection, remember to also revoke the authorization on the external service's connected-apps settings. This ensures a complete cleanup and prevents stale tokens from lingering on the provider side.
Go to the MCP Hub (Connections page). Pick an active connection and open it to inspect its tool list. Then navigate to the Execution Log and filter by that connection's tools — review the origin column to see which OUPI surfaces have been invoking it.
A single MCP connection propagates across every OUPI surface — ChatPro badges, Studio agents, workflows, RAG imports, and Deep Research — with zero per-product setup. Each connection exposes a transparent tool list (what the AI actually sees), and every invocation is logged with tool, time, origin, and outcome. Costs fold into the triggering operation's credits. Broken connections almost always mean expired OAuth — re-authorize in the Hub and revoke on the service side for clean teardown.