Support different transports in Client

[Kludex]

This pull request updates the MCP client codebase to simplify and standardize the transport interface for client-server communication. The main change is that the streamable_http_client and related transports now return only a 2-tuple (read_stream, write_stream), removing the previously returned get_session_id callback. The documentation and all usage examples have been updated accordingly. Additionally, a new Transport protocol is introduced to formalize the transport interface, and the in-memory transport is refactored to support async context management. Minor documentation clarifications and type cleanups are also included.

Key changes:

Transport Interface Simplification

• The streamable_http_client now returns a 2-tuple (read_stream, write_stream) instead of a 3-tuple with get_session_id. All usages in code, examples, and documentation have been updated to match this new interface. [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17]

• The get_session_id callback is now documented as removed, with migration instructions and alternatives using httpx event hooks provided in docs/migration.md.

Transport Protocol and In-Memory Transport Refactor

• Introduced a new Transport protocol in src/mcp/client/_transport.py to standardize the interface for all client transports. [1] [2]

• Refactored InMemoryTransport to implement async context management (__aenter__/__aexit__), and updated its usage and documentation for consistency with the new protocol. [1] [2] [3] [4]

Documentation and Typing Improvements

• Updated README.v2.md, and all relevant example clients to reflect the new transport interface and provide clear, accurate usage patterns. [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12]

• Minor documentation clarifications and type annotation cleanups, including a note in CLAUDE.md about import placement and type signature simplifications. [1] [2]

These changes improve consistency, usability, and maintainability of the MCP client transport layer and its documentation.

[claude]

Code review

No issues found. Checked for bugs and CLAUDE.md compliance.

[claude]

Code review

No issues found. Checked for bugs and CLAUDE.md compliance.

[Copilot]

Pull request overview

This PR introduces support for different transports in the Client class, making it more flexible and extensible. The changes enable the Client to work with URL strings, Transport instances, or Server/MCPServer instances, while standardizing the transport interface.

Changes:

• Introduces a Transport protocol that standardizes how client transports work
• Refactors Client class from manual init to dataclass for better maintainability
• Removes the get_session_id callback from streamable_http_client to align with Transport protocol (now returns 2-tuple instead of 3-tuple)
• Updates InMemoryTransport to implement the Transport protocol as an async context manager
• Updates all tests, examples, and documentation to reflect the 2-tuple return signature

Reviewed changes

Copilot reviewed 25 out of 25 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
src/mcp/client/_transport.py	New file defining the Transport protocol and TransportStreams type
src/mcp/client/_memory.py	Refactored InMemoryTransport to implement Transport protocol via aenter/aexit
src/mcp/client/client.py	Converted Client to dataclass, added support for URL strings and Transport instances
src/mcp/client/streamable_http.py	Removed get_session_id callback, now returns 2-tuple matching Transport protocol
src/mcp/client/session_group.py	Updated to unpack 2-tuple from streamable_http_client
src/mcp/client/__init__.py	Added Transport to public exports
tests/*	Updated all tests to unpack 2-tuple instead of 3-tuple
examples/*	Updated all example clients to use new 2-tuple signature
docs/migration.md	Added migration guide for get_session_id callback removal
README.md, README.v2.md	Updated code examples to reflect new signature
.github/actions/conformance/client.py	Updated conformance tests for new signature
CLAUDE.md	Emphasized rule about imports at top of file

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[felixweinberger]

I'm curious about the motivation for this.

Doesn't this leak _prefixed arguments to the public API? E.g. you could now do:

Client(server, _session=some_session, _exit_stack=some_stack) 

And I think it'll also appear in repr() now. Which feels odd / potentially opening up the API for people to do weird stuff that they shouldn't do. Or maybe it's a case of having the _prefix be the message "you're doing weird stuff" and have that be enough?

Not fundamentally opposed - just wondering is the the primary motivation to simplify the class definition here so we avoid self._foo = foo boilerplate?

[Kludex]

Not fundamentally opposed - just wondering is the the primary motivation to simplify the class definition here so we avoid self._foo = foo boilerplate?

Yes. And some other stuff like documenting the parameter close to the definition. And the post_init is focused on the real logic.

[felixweinberger]

Looks great, thanks!