Academy
4. Inspecting Exposed Tools
Open a live connection to examine the exact tool list the AI sees, understand each tool's described action, and verify that granted scopes match your intent.
When you create an MCP connection, the service exposes a set of tools — discrete actions the AI can invoke (e.g., "create_issue", "list_emails", "search_contacts"). This tool list is not abstract documentation; it is the exact interface the AI model sees at runtime when it decides whether and how to call the service. Understanding this list is critical because any tool present is a tool the AI may call autonomously. If a tool you didn't intend to expose is listed, the AI could still invoke it within the granted scopes. Inspecting tools is therefore a security and correctness step, not just a curiosity.
To inspect tools, navigate to your Connections page and open the connection you want to audit. The detail view lists every tool the connection exposes, each with a description of the action it performs. These descriptions are what the AI reads to decide which tool fits a user's request, so they also tell you exactly what the AI believes each tool does. Compare this list against the scopes you granted during OAuth: the tools should be a subset of what those scopes authorize. If you see tools that exceed your intent — e.g., write-access tools when you only wanted reads — the fix is to delete the connection and recreate it with narrower scopes on the service side.
Tool inspection is also your first debugging step. If the AI never calls a service you connected, open the connection and check whether the expected tools actually appear. A missing tool usually means the granted scopes were too narrow or the service didn't expose that capability via MCP.
Because connections work platform-wide — ChatPro, OUPI Studio, the Assistant, workflow steps, RAG imports, and Deep Research — every tool you see in the connection detail is available in all of those contexts. There is no per-context filtering. This means a tool you expose for a ChatPro conversation is equally callable from a workflow step. Keep this in mind when auditing: if a tool is too powerful for unattended workflow use, it may be too powerful to leave connected at all.
Open the MCP Hub → Connections page. Pick a live connection and open it. Read through every exposed tool and its description. Cross-check: does each tool match a scope you intentionally granted? Then open the Execution log to see which of those tools the AI has actually called, and verify the outcomes match your expectations.
After inspecting tools, make it a habit to check the execution log regularly. Every tool call is recorded with the tool name, timestamp, origin (conversation, workflow, etc.), and outcome. This is your audit trail — use it to confirm the AI is only doing what you intended.
Inspecting exposed tools means opening a connection's detail to see the exact tool list the AI uses at runtime. Each tool's description tells you what the AI believes it can do. Compare tools against your granted OAuth scopes to ensure nothing exceeds your intent. Remember that connections are platform-wide — every tool is available everywhere. Use the execution log as your ongoing audit trail to verify actual behavior matches your design.