[#36] feat: add HTTP client transport support by mchades · Pull Request #37 · agenticdataprotocol/python-sdk

mchades · 2026-04-15T09:57:50Z

What changes were proposed in this pull request?

Add HttpClientTransport for communicating with ADP servers over HTTP, enabling connections to remote or already-running servers without managing a subprocess.

Changes:

Add httpx>=0.27.1 as a required dependency (following MCP SDK precedent)
Implement HttpClientTransport in src/adp_sdk/transports/http/ (POST-based JSON-RPC request/response)
Add http_client() async context manager convenience function in src/adp_sdk/__init__.py
Update src/adp_sdk/transports/__init__.py to re-export HttpClientTransport
Add 17 unit tests covering lifecycle, messages, and error responses
Update README.md with HTTP transport documentation and examples

Why are the changes needed?

The adp-hypervisor is adding HTTP transport support (adp-hypervisor#89). The SDK needs a corresponding client-side HTTP transport so that clients can connect to ADP servers running in HTTP mode.

Fix: #36

Does this PR introduce any user-facing change?

Yes:

New public class HttpClientTransport for HTTP-based communication.
New convenience function http_client() — async context manager that creates an initialized ClientSession over HTTP.
New dependency: httpx>=0.27.1 is now a required dependency.

How was this patch tested?

17 new unit tests in tests/test_http_client.py covering:
- Transport lifecycle (start/close, context manager, error states)
- Message send/receive (write/read pairing, buffer clearing, multiple cycles)
- HTTP headers (Content-Type, custom headers, Authorization)
- Error responses (HTTP 400/415/500 with JSON-RPC error bodies)
All existing tests continue to pass (114/114, excluding pre-existing jsonschema issue)
ruff check ✅, ruff format ✅, pyright 0 errors ✅

Add HttpClientTransport for communicating with ADP servers over HTTP, enabling connections to remote or already-running servers without managing a subprocess. Changes: - Add httpx>=0.27.1 as a required dependency - Implement HttpClientTransport (POST-based JSON-RPC request/response) - Add http_client() async context manager convenience function - Add 17 unit tests covering lifecycle, messages, and error responses - Update README.md with HTTP transport documentation and examples Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

Pull request overview

Adds an HTTP-based client transport to the ADP Python SDK so clients can talk to remote/already-running ADP servers without spawning a subprocess.

Changes:

Introduces HttpClientTransport (POST-based JSON-RPC) and re-exports it from the transports package.
Adds http_client() async context manager for creating an initialized ClientSession over HTTP.
Updates dependencies/docs and adds a dedicated unit test suite for the new transport.

Reviewed changes

Copilot reviewed 7 out of 9 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
uv.lock	Moves `httpx` into required deps and bumps lock revision/version metadata.
pyproject.toml	Adds `httpx>=0.27.1` as a required dependency.
src/adp_sdk/transports/http/client.py	Implements `HttpClientTransport` with request/response buffering semantics.
src/adp_sdk/transports/http/init.py	Exposes `HttpClientTransport` from the `http` transport package.
src/adp_sdk/transports/init.py	Re-exports `HttpClientTransport` at `adp_sdk.transports`.
src/adp_sdk/init.py	Adds `http_client()` convenience context manager and exports `HttpClientTransport`.
tests/test_http_client.py	Adds unit tests for HTTP transport lifecycle, headers, message flow, and error bodies.
README.md	Documents HTTP transport usage alongside stdio transport.
.gitignore	Ignores JetBrains `.idea/` directory.

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

Copilot · 2026-04-21T09:50:51Z

+        authorization: Optional authorization header value for requests
+            (e.g. ``basic_auth("user", "pass")``).  When provided, it is
+            sent as the HTTP ``Authorization`` header so the server can
+            extract it at the transport level.
+
+    Yields:
+        An initialized :class:`ClientSession`.
+    """
+    effective_headers = dict(headers) if headers else {}
+    if authorization is not None:


http_client() silently overwrites any user-provided Authorization header when both headers (with an Authorization key) and the authorization argument are provided. Consider either (a) raising a ValueError when both are set, or (b) documenting/implementing a clear precedence rule (e.g., only set the header if it’s not already present).

Suggested change

authorization: Optional authorization header value for requests

(e.g. ``basic_auth("user", "pass")``). When provided, it is

sent as the HTTP ``Authorization`` header so the server can

extract it at the transport level.

Yields:

An initialized :class:`ClientSession`.

"""

effective_headers = dict(headers) if headers else {}

if authorization is not None:

Must not include ``Authorization`` when ``authorization`` is also

provided.

authorization: Optional authorization header value for requests

(e.g. ``basic_auth("user", "pass")``). When provided, it is sent

as the HTTP ``Authorization`` header so the server can extract it

at the transport level. Must not be combined with an

``Authorization`` entry in ``headers``.

Yields:

An initialized :class:`ClientSession`.

Raises:

ValueError: If both ``authorization`` and ``headers["Authorization"]``

are provided.

"""

effective_headers = dict(headers) if headers else {}

if authorization is not None:

if "Authorization" in effective_headers:

raise ValueError(

"Provide either headers['Authorization'] or authorization, not both."

)

Fixed in 22b39de. http_client() now raises ValueError when authorization= is combined with an existing Authorization header in headers, and the check is case-insensitive. I also updated the docstring and added a regression test for this path.

Copilot · 2026-04-21T09:50:51Z

+        headers = {**self._headers, "Content-Type": "application/json"}
+        response = await self._client.post(self._url, content=message, headers=headers)
+
+        self._pending_response = response.text
+        logger.debug("HTTP %s %s -> %d", "POST", self._url, response.status_code)


write_message() will overwrite _pending_response if it’s called twice without an intervening read_message(), silently dropping the previous HTTP response. Since this transport is explicitly request/response paired, consider raising a RuntimeError (or similar) when _pending_response is still set to enforce correct sequencing and avoid hard-to-debug response/id mismatches when HttpClientTransport is used directly (outside ClientSession).

Fixed in 22b39de. HttpClientTransport.write_message() now raises RuntimeError if the previous response has not been consumed yet, so we no longer overwrite _pending_response. Added a regression test for direct transport usage as well.

Where did you fix the things?

This should not happen through ClientSession, since ClientSession._send_request() serializes calls with _send_lock and always performs write_message() followed immediately by read_message().

The new RuntimeError only triggers when HttpClientTransport is used directly with an invalid write -> write sequence while the previous response is still buffered in _pending_response.

src/adp_sdk/transports/http/client.py:103-106 adds that guard.

tests/test_http_client.py:241-247 adds the regression test for the direct transport usage path.

I mean where is the commit id 22b39de, I don't see this in the PR.

You were right — 22b39de had been pushed to the wrong remote earlier. I have now pushed it to the PR branch (mchades/python-sdk:http-transport), so that commit is part of this PR now.

jerryshao · 2026-04-21T09:56:12Z



+@asynccontextmanager
+async def http_client(


Do we have the concurrency requirement? This client seems can only work in a single thread.

Current HttpClientTransport is intentionally a single in-flight request/response transport, which matches the existing SDK transport contract.

Transport already documents that read_message() / write_message() are not concurrent-safe, and ClientSession serializes concurrent requests with an internal asyncio.Lock() to prevent response mismatches.

So this PR keeps the current client model consistent rather than introducing a new concurrency limitation specific to HTTP. If we want true multiple in-flight HTTP requests, that would likely require a broader transport/session design change rather than a small adjustment in this PR.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

mchades requested a review from jerryshao April 15, 2026 10:00

jerryshao requested a review from Copilot April 21, 2026 09:46

Copilot started reviewing on behalf of jerryshao April 21, 2026 09:47 View session

Copilot AI reviewed Apr 21, 2026

View reviewed changes

jerryshao reviewed Apr 21, 2026

View reviewed changes

fix: guard HTTP client request sequencing

22b39de

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

mchades requested a review from jerryshao April 22, 2026 06:33

jerryshao approved these changes Apr 22, 2026

View reviewed changes

jerryshao merged commit f390b67 into agenticdataprotocol:main Apr 22, 2026
3 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[#36] feat: add HTTP client transport support#37

[#36] feat: add HTTP client transport support#37
jerryshao merged 2 commits into
agenticdataprotocol:mainfrom
mchades:http-transport

mchades commented Apr 15, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Apr 21, 2026

Uh oh!

mchades Apr 21, 2026

Uh oh!

Copilot AI Apr 21, 2026

Uh oh!

mchades Apr 21, 2026

Uh oh!

jerryshao Apr 22, 2026

Uh oh!

mchades Apr 22, 2026

Uh oh!

jerryshao Apr 22, 2026

Uh oh!

mchades Apr 22, 2026

Uh oh!

jerryshao Apr 21, 2026 •

edited

Loading

Uh oh!

mchades Apr 21, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

mchades commented Apr 15, 2026

What changes were proposed in this pull request?

Why are the changes needed?

Does this PR introduce any user-facing change?

How was this patch tested?

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Apr 21, 2026

Choose a reason for hiding this comment

Uh oh!

mchades Apr 21, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 21, 2026

Choose a reason for hiding this comment

Uh oh!

mchades Apr 21, 2026

Choose a reason for hiding this comment

Uh oh!

jerryshao Apr 22, 2026

Choose a reason for hiding this comment

Uh oh!

mchades Apr 22, 2026

Choose a reason for hiding this comment

Uh oh!

jerryshao Apr 22, 2026

Choose a reason for hiding this comment

Uh oh!

mchades Apr 22, 2026

Choose a reason for hiding this comment

Uh oh!

jerryshao Apr 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mchades Apr 21, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

jerryshao Apr 21, 2026 •

edited

Loading