Summary

In some cases we need to save the session id or to restore an existing session id and to continue from it, in some cases we need to move a session across process boundary (for example debug or human in the loop scenarios).

The transport previously owned session state directly via self.session_id, making it impossible for consumers to observe or persist session IDs externally (e.g. persisting to a debug state endpoint for playground/agent resume scenarios).

This PR introduces a SessionInfo base class with async get_session_id / set_session_id methods that the transport delegates all session ID storage to. Consumers can subclass SessionInfo to add side-effects like HTTP persistence, and pass it via the new session_info parameter on StreamableHTTPTransport and streamable_http_client.

Key changes

• Add SessionInfo class — base class with async get/set methods; default implementation stores in a plain attribute (backwards compatible)
• Transport delegates to _session_info — replaces direct self.session_id usage
• _prepare_headers and _maybe_extract_session_id_from_response are now async — to support subclasses that need I/O (e.g. loading from a remote store)
• Remove unused session_id field from RequestContext — headers are built from _prepare_headers, not from the context
• streamable_http_client accepts optional session_info parameter — passed through to the transport constructor
• get_session_id is now async — delegates to SessionInfo.get_session_id()

Backwards compatibility

• Default behavior is unchanged: when no session_info is provided, a plain SessionInfo() is created internally
• All 49 existing streamable HTTP tests pass without modification
• The removed RequestContext.session_id field was not used by any handler (headers come from _prepare_headers)

Example usage

Track the session ID externally:

session_info = SessionInfo()

async with streamable_http_client(url, http_client=client, session_info=session_info) as (read, write):
    session = ClientSession(read, write)
    await session.initialize()
    # session_info.session_id now holds the server-assigned session ID
    print(f"Session ID: {await session_info.get_session_id()}")

Resume an existing session (skip initialize):

# Reuse a session ID from a previous connection
session_info = SessionInfo(session_id="previously-saved-session-id")

async with streamable_http_client(url, http_client=client, session_info=session_info) as (read, write):
    session = ClientSession(read, write)
    # No initialize() call — the transport includes the existing session ID in headers
    tools = await session.list_tools()