Skip to main content
Studio ships with built-in tools for terminals, diagnostics, inventory, diagrams, and search — everything Copilot needs for the core operational loop. When you need Copilot to reach your own systems, you add connectors or plug in an MCP server. Connectors are named REST or SOAP endpoints with authentication. You configure them inside Studio, and Copilot treats each endpoint as a callable action with its own description and approval. MCP servers are processes you run elsewhere that expose tools over a standard protocol — when you connect one, its tools appear in the Copilot catalog alongside the built-ins. Both show up the same way in a conversation: Copilot explains what it’s about to do, shows the request, and asks for approval before sending.

When to use which extension

Use a connector when the system has a normal API and you want Studio to own the endpoint shape. Use MCP when another service already exposes tools, or when you want a custom tool process to evolve outside Studio.
NeedUseWhy
Call an internal REST endpoint.ConnectorStudio can describe the endpoint, auth, parameters, and response schema directly.
Call a SOAP service.ConnectorThe request and response can be wrapped as a named operation Copilot can understand.
Expose several tools from a local or hosted service.MCP serverThe server owns the tool catalog and Studio refreshes it.
Let Copilot query a monitoring system.Connector or MCPPick connector for a few known endpoints; pick MCP for a richer tool suite.
Trigger changes in an external system.Connector with approvalsKeep the request visible and reviewable before anything is sent.

Connectors

A connector is a named integration with your API. You configure it once, define the endpoints you want Copilot to be able to call, and the result is a set of typed actions Copilot can pick from when the conversation asks for them.

Creating a connector

  • Name and description. What the connector is for, in your words.
  • Base URL. The root URL every endpoint hangs off.
  • Authentication. None, basic, bearer token, or a custom header — pick whatever your API uses.
  • Named endpoints. For each operation you want to expose: method, path, query, and body schema. Each endpoint gets its own description and its own approval.

Endpoint design

Good endpoint descriptions make Copilot safer. Name the operation the way an operator would ask for it, then describe the inputs, side effects, and expected response.
FieldGuidance
NameUse verbs: listAlarms, createTicket, getInterfaceStats.
DescriptionSay what the endpoint does and when to use it.
ParametersPrefer explicit fields over a single free-form blob.
Body schemaInclude required fields, enums, and examples where possible.
Response schemaDescribe the shape Copilot should expect so it can table or summarize results.
Side effectsState whether the endpoint is read-only or changes external state.
Start read-only. Add write endpoints only after the connector has proven useful and the approval prompt gives operators enough information to judge the call.

Calling a connector from Copilot

You ask in natural language. “List current alarms on the monitoring connector.” “Create a ticket for this host.” Copilot picks the endpoint that matches, fills in the parameters, shows you the request, and asks for approval before sending. The response becomes evidence in the conversation — you can quote it, table it, or feed it into a procedure.

Approval posture

Connector calls require approval by default. External systems can take real actions, and the approval is the one place where you see exactly what Copilot is about to send. Moving a connector to auto-approved is a per-connector choice — appropriate for read-only endpoints you trust, less so for anything that changes state. For write endpoints, include the external target in the endpoint description and request body: ticket queue, monitoring tenant, customer ID, site, or device. A request that says exactly where it is going is easier to approve safely.

MCP servers

An MCP server is a process that exposes tools over a standard protocol. You run or subscribe to the server somewhere, point Studio at it, and its tools show up in the Copilot tool catalog. Studio supports HTTP and SSE transports. To connect one you provide:
  • The server URL.
  • Auth — a bearer token or a custom header.
  • An optional name used as a prefix for the server’s tools.
Studio periodically refreshes the tool list from the server. If a tool disappears from the server, Copilot stops offering it. If the server disconnects, Copilot stops calling into it until the connection is back. Treat MCP tools as production code. Version them, describe their inputs clearly, and keep destructive operations narrow. Studio can gate calls with the same approval model as built-in tools, but the tool description is what helps Copilot choose the right operation in the first place.

Built-in MCP tools

Studio itself provides the tool catalog Copilot uses every day — terminals, diagnostics, inventory, parsing, diagrams, search. You don’t configure these; they’re always available. They’re gated by the same approval posture you pick in Copilot, which means the same safety model covers your built-ins, your connectors, and any MCP servers you add.

Troubleshooting connectors

SymptomFix
Calls fail with 401Re-check the auth configuration; rotate the token if needed.
Calls fail with 403Check the account’s permissions on the target API.
Server returns unexpected shapeAdd or adjust the endpoint’s response schema so Copilot can parse.
MCP server disconnectsCheck network path, auth token, and server health.
Copilot picks the wrong endpointRename or rewrite endpoint descriptions so the intended use is clearer.
Approval prompt is hard to reviewMake parameters more explicit and include a dry-run or summary endpoint.

AI Copilot

How Copilot discovers and calls connector and MCP tools.

Security and privacy

Where credentials live and how approvals gate external calls.