Model Context Protocol

Connect and test MCP tools

Connect an MCP server, refresh its tool catalog, inspect an input schema, run a test call, and disconnect safely.

Connect a configured Model Context Protocol (MCP) server and prove that one of its tools works with a controlled test call. Use this workflow after adding a server, changing its endpoint or credentials, or investigating a stale tool catalog.

Before you begin

You must be signed in and have an MCP server listed under CONFIGURED MCP SERVERS. For setup choices and transport requirements, see Manage MCP servers.

Choose a read-only tool for the first test. A tool marked WRITE can change data in the connected service, and a tool marked HITL requires human approval. Prepare synthetic input that matches the tool's schema; do not put secrets into test arguments.

ℹ️The connection, discovery, and test actions apply only to a server owned by your signed-in account. A missing or inaccessible server cannot be operated through these controls.

Connect and inspect tools

Connect the server

Open the MCP area from the Tool Registry. Under CONFIGURED MCP SERVERS, find the server and select CONNECT.

Wait for the server status to become CONNECTED. Agent Studio contacts the configured SSE endpoint or starts the configured stdio command, then caches the tools reported by the server. If the connection fails, check the endpoint, command, and required credentials before continuing.

Refresh tool discovery

Expand the connected server and run Rediscover when you need a fresh catalog. This action queries the server again and replaces the cached tool definitions.

The DISCOVERED TOOLS count updates when discovery completes. If no tools appear, the inspector explains that the server must be connected before it can query the remote tools/list operation.

Inspect the input schema

Select TOOLS, then select the name of the tool you intend to call. Agent Studio expands the tool's JSON input schema below its description.

Review the property names, types, and required fields. Also check the READ, WRITE, and HITL indicators before choosing test data. The schema, rather than a copied example from another tool, is the authority for the arguments this server accepts.

Prepare test arguments

Select TEST TOOL. In CALL ARGUMENTS (JSON PAYLOAD), replace the generated sample with valid JSON that satisfies the displayed schema. For a hypothetical read-only ticket lookup, the payload might be:

json

{
  "ticketId": "TICKET-1042"
}

Use [ FORMAT JSON ] to normalize valid JSON or [ RESET SAMPLE ] to restore the generated payload. Agent Studio reports Arguments must be valid JSON locally when the editor content cannot be parsed.

Execute the call

Select EXECUTE MCP CALL. Agent Studio sends the selected tool name and the JSON object to the connected server.

A successful call shows STATUS: SUCCESS (200 OK), the duration, and RESPONSE STREAM OUTPUT. An unsuccessful call shows STATUS: EXECUTION ERROR and the returned error. Inspect the result before using the tool in an agent graph.

Disconnect when finished

Close the test console, then select DISCONNECT on the server when it should no longer remain connected. The server status changes to DISCONNECTED; the saved server configuration remains available for a later connection.

Verify the result

✅The server shows CONNECTED during the test, DISCOVERED TOOLS contains the selected tool, and the test console reports STATUS: SUCCESS (200 OK) with output appropriate for the synthetic input. After cleanup, the server shows DISCONNECTED.

Troubleshooting

Next, use advanced MCP operations to work with resources, prompts, sampling, progress, and server health.