Test MCP server discovery and tool calls locally

Goal

Verify the MCP interface in isolation without depending on production services or accounts.

Steps

1. Create a local test-server configuration that uses no real account or production data. [mcp-test] [mcp-tools]
2. Create an in-memory session with async with Client(mcp, raise_exceptions=True) as client:; no subprocess or port is needed. [mcp-test] [mcp-tools]
3. Inspect input_schema from client.list_tools(), then invoke a read-only test tool with await client.call_tool(name, args). [mcp-test] [mcp-tools]
4. Submit missing-field, wrong-type, and unknown-tool requests and verify controlled protocol errors. [mcp-test] [mcp-tools]
5. Close the session and confirm child processes and temporary resources are released. [mcp-test] [mcp-tools]

How to verify

Local tests cover initialization, capability discovery, valid and invalid calls, and shutdown without contacting production.

Troubleshooting

If initialization fails, compare client/server protocol versions and inspect the controlled error without enabling production access.