A developer wants to extend a Copilot Studio agent with tools hosted on an external MCP (Model Context Protocol) server. The MCP server exposes a weather lookup tool and a calendar scheduling tool. To register this MCP server in Copilot Studio, what information must the developer provide during configuration?
Select an answer to reveal the explanation.
![Portrait infographic 1536x2048 titled 'Registering an MCP Server in Copilot Studio'. Show a registration flow: Copilot Studio MCP Settings panel with two input fields: 'Server Endpoint URL' (https://mcp.example.com/api) and 'Authentication' (OAuth 2.0 credentials). Arrow labeled 'Auto-Discovery Request' points to an MCP Server box showing Tool List response: [Weather Lookup, Calendar Scheduling]. Arrow labeled 'Tools Available in Agent' points to agent's Action palette showing both tools. Side panel shows the MCP protocol spec symbol. Microsoft Fluent design, teal/blue palette, clean technical diagram with labeled callouts.](/media/microsoft/ab-620/AB-620-e5-q006.webp)
Short Explanation
Registering an MCP server is like giving a new contractor your office address and access badge—you give Copilot Studio the server's endpoint URL and any required credentials, then it discovers the tools automatically.
Full Explanation
Model Context Protocol (MCP) is an open standard that allows AI agents to connect to external tool servers that expose capabilities such as data lookups, API calls, and integrations. In Copilot Studio, registering an MCP server requires providing the server's endpoint URL, typically an HTTPS endpoint conforming to the MCP specification, and any authentication credentials the server requires, such as an OAuth 2.0 client ID and secret or an API key. Once registered, Copilot Studio queries the MCP server's tool discovery endpoint to automatically enumerate available tools, so the developer does not need to manually define each tool.
The configuration that provides the MCP server endpoint URL and, if the server requires authentication, the OAuth 2.0 or API key credentials so Copilot Studio can discover and invoke the server's tools is correct. This matches the MCP specification's server-side tool discovery model: the agent connects to the endpoint, authenticates, and the server returns a list of available tools with their schemas. Copilot Studio then makes those tools available in the agent's action configuration.
The configuration that gives only a REST API base URL and an API key stored in Azure Key Vault, with no additional configuration needed, is wrong because a base URL alone is not sufficient for secure MCP servers. Authentication credentials are required when the server requires them, and the URL must conform to the MCP protocol endpoint structure rather than being treated as a generic REST base URL with no further setup.
The configuration that names each individual tool exposed by the MCP server and configures them as separate HTTP connectors in Power Platform is wrong because MCP tools are not configured as separate Power Platform HTTP connectors. MCP is a distinct protocol from Power Platform custom connectors. Treating MCP tools as HTTP connectors would bypass MCP's tool discovery mechanism and require manual schema definition for each tool.
The configuration that requires a JSON manifest file listing all tool names and parameters, uploaded directly to the Copilot Studio maker portal, is wrong because Copilot Studio does not require the developer to upload a JSON manifest manually for MCP registration. The MCP server itself exposes a discovery endpoint that Copilot Studio queries automatically after registration. Manual manifest upload is a pattern used with OpenAPI specifications for custom connectors, not with MCP servers.
Exam tip: MCP server registration in Copilot Studio follows a two-field pattern: endpoint URL plus authentication. After that, tool discovery is automatic. Don't confuse MCP registration with custom connector registration, which does require manual OpenAPI spec upload.