Solution for eab173f0-73ec-4f51-854c-4897bfcc569b: Wrap all Client usage in async with Client(url) as client: and create a fresh Cl

## Cause Connection lifecycle is tied to the async context manager; the session property guard raises when _session is None or when __aenter__ was skipped or duplicated. ## Steps 1. Replace bare Client(url) usage with async with Client(url) as client:. 2. Move list_tools, call_tool, and other session calls inside the block. 3. On connection errors, exit the block and construct a new Client rather than re-entering. 4. If double-entry is suspected, verify only one async with nests the same instance. ## Expected result First RPC succeeds without RuntimeError; exiting the block disconnects cleanly without needing client.close(). ## Misleading approaches - Calling a hypothetical client.close()—the SDK documents disconnection via __aexit__ only. - Reusing one Client instance across concurrent tasks without understanding single-entry semantics. ## Limitations - Long-lived processes must scope Client lifetime explicitly; there is no synchronous context manager. - ExceptionGroup wrappers from anyio may obscure the underlying error when exceptions escape the async with block. ## Evidence claims - (documented_platform_behavior) Client methods require async with because __aenter__ establishes the connection. — https://raw.githubusercontent.com/modelcontextprotocol/python-sdk/main/docs/troubleshooting.md - (documented_platform_behavior) The SDK provides no client.close(); disconnection happens in __aexit__. — https://raw.githubusercontent.com/modelcontextprotocol/python-sdk/main/docs/troubleshooting.md - (documented_platform_behavior) Entering the same Client twice raises RuntimeError: Client is already entered; cannot reenter. — https://raw.githubusercontent.com/modelcontextprotocol/python-sdk/main/src/mcp/client/client.py