feat(sigv4_helper): inject AWS_REGION in _meta #62

kyoncal · 2025-10-30T14:42:40Z

Summary

Implemented automatic AWS_REGION metadata injection into MCP JSON-RPC requests using httpx request hooks. Added _inject_metadata_hook function to sigv4_helper.py that injects metadata into the _meta field of request params, merging with existing metadata when present (existing values take precedence). Includes unit tests and an end-to-end integration test with a new echo_metadata tool in the test MCP server to verify metadata transmission.

User experience

AWS_REGION is now automatically injected into all MCP requests transparently - no manual configuration required. The proxy adds region information to the _meta field, making it available to MCP servers without any client-side code changes.

Checklist

If your change doesn't seem to apply, please leave them unchecked.

I have reviewed the contributing guidelines
I have performed a self-review of this change
Changes have been tested
Changes are documented

Is this a breaking change? (Y/N)

Yes
No

Please add details about how this change was tested.

Did integration tests succeed? (Only locally, ACR doesn't work for me on this commit or main head)
If the feature is a new use case, is it necessary to add a new integration test case?

Acknowledgment

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

arangatang

Thanks for the review, glad to see it working! Some comments but nothing fundamental.

arangatang · 2025-10-30T15:16:29Z

mcp_proxy_for_aws/sigv4_helper.py

+                # Ensure params exists
+                # if 'params' not in body:
+                #     body['params'] = {}


Should this be commented out?

I believe so. There is no MCP request that shouldn't contains a params field.

Then either delete it, or uncomment it. I'd leave it there, fix it and maybe log a warning if this is expected?

arangatang · 2025-10-30T15:17:12Z

mcp_proxy_for_aws/sigv4_helper.py

+                # Create new content
+                new_content = json.dumps(body).encode('utf-8')
+
+                from httpx._content import ByteStream


Lets not import in the middle of a function.

Agreed. Will fix.

arangatang · 2025-10-30T15:18:59Z

mcp_proxy_for_aws/sigv4_helper.py

+
+        except (json.JSONDecodeError, KeyError, TypeError) as e:
+            # Not a JSON request or invalid format, skip metadata injection
+            logger.debug('Skipping metadata injection: %s', e)


Lets make this error level.

arangatang · 2025-10-30T15:24:08Z

mcp_proxy_for_aws/sigv4_helper.py

+                if '_meta' not in body['params']:
+                    body['params']['_meta'] = {}
+                else:
+                    print()


Why this print?

oops. remnant from debugging.

arangatang · 2025-10-30T15:27:39Z

mcp_proxy_for_aws/sigv4_helper.py

+                existing_meta = body['params']['_meta']
+
+                # Merge metadata (existing takes precedence)
+                if isinstance(existing_meta, dict):
+                    body['params']['_meta'] = {**metadata, **existing_meta}
+                else:
+                    body['params']['_meta'] = metadata


nit: I think this can be replaced with

body['params']['_meta'] = metadata | body['params']['_meta']

arangatang · 2025-10-30T15:36:38Z

tests/unit/test_sigv4_metadata_injection.py

+        await _inject_metadata_hook(metadata, request)
+
+        # Clear cached content and read from the new stream
+        del request._content


Why the delete here?

~~After looking at the implementation of request.aread(), I've reached the conclusion that it is indeed unnecessary.~~ I've moved it to sigv4_helper.py, because it is necessary. Since _content functions as a cache, it seems asthough the cached value is read instead of the updated one. Although it doesn't make sense for the caller to need to do this.

arangatang · 2025-10-30T15:38:32Z

mcp_proxy_for_aws/sigv4_helper.py

+
+                # Merge metadata (existing takes precedence)
+                if isinstance(existing_meta, dict):
+                    body['params']['_meta'] = {**metadata, **existing_meta}


I think we may need to log if there are failed overwrites here, i.e. if a key was available in both, then this would make it clearer what happened.

arangatang · 2025-10-30T15:41:54Z

tests/integ/test_proxy_simple_mcp_server.py

+
+
+@pytest.mark.asyncio(loop_scope='module')
+async def test_metadata_injection_aws_region(


Thanks for the integ test!

arangatang · 2025-10-30T15:43:11Z

tests/unit/test_sigv4_metadata_injection.py

+        assert modified_body['params']['_meta']['field2'] == 'original'
+        assert modified_body['params']['_meta']['AWS_REGION'] == 'us-west-2'
+
+    @pytest.mark.asyncio


Duplicate decorator.

kyoncal · 2025-11-03T10:49:34Z

Turn this PR to a draft, as it breaks Sigv4 auth.

…ta_hook to allow for proper resigning of sigv4 to work

arangatang · 2025-11-04T12:10:45Z

mcp_proxy_for_aws/sigv4_helper.py

            raise e


+def _resign_request_with_sigv4(


Tbh, if we go with this approach we should not sign it at all before this point and only do signing here. I.e. no need to re-sign.

arangatang · 2025-11-04T12:15:05Z

mcp_proxy_for_aws/sigv4_helper.py

+    # Set the new Content-Length
+    request.headers['Content-Length'] = str(len(request.content))
+
+    logger.info('Headers after cleanup: %s', request.headers)


Don't think we should log this anymore, atleast not with info level.

arangatang · 2025-11-04T12:15:50Z

mcp_proxy_for_aws/sigv4_helper.py

+    service: str,
+    profile: Optional[str] = None,
+) -> None:
+    """Re-sign an HTTP request with AWS SigV4 after content modification.


We should probably only do signing here and not re-sign. But this changes the core way the proxy works. We need to align on this.

arangatang · 2025-11-04T12:16:07Z

mcp_proxy_for_aws/sigv4_helper.py

+    # Get AWS credentials
+    session = create_aws_session(profile)
+    credentials = session.get_credentials()
+    logger.info('Re-signing request with credentials for access key: %s', credentials.access_key)


This should not be logged / be on debug level.

arangatang · 2025-11-04T12:16:26Z

mcp_proxy_for_aws/sigv4_helper.py

+    )
+
+    # Sign the request
+    logger.info('AWS request before signing: %s', aws_request.headers)


Same as above, should be debug

arangatang · 2025-11-04T12:16:32Z

mcp_proxy_for_aws/sigv4_helper.py

+    # Sign the request
+    logger.info('AWS request before signing: %s', aws_request.headers)
+    signer.add_auth(aws_request)
+    logger.info('AWS request after signing: %s', aws_request.headers)


Same as above, should be debug

arangatang · 2025-11-04T12:16:36Z

mcp_proxy_for_aws/sigv4_helper.py

+
+    # Update request headers with signed headers
+    request.headers.update(dict(aws_request.headers))
+    logger.info('Request headers after re-signing: %s', request.headers)


Same as above, should be debug

arangatang · 2025-11-04T12:18:24Z

mcp_proxy_for_aws/sigv4_helper.py

+                            )
+                    body['params']['_meta'] = {**metadata, **existing_meta}
+                else:
+                    logger.info('Replacing non-dict _meta value with injected metadata')


What do you mean with non-dict here?
Maybe:

'No conclicting _meta entries exist.`

arangatang · 2025-11-04T12:19:58Z

mcp_proxy_for_aws/sigv4_helper.py

+                # Re-sign the request with the new content
+                _resign_request_with_sigv4(request, region, service)
+
+                logger.info('Injected metadata into _meta: %s', body['params']['_meta'])


nit: MaybeUpdated _meta after injection of additional _meta parameters: <meta>

kyoncal requested a review from a team as a code owner October 30, 2025 14:42

kyoncal requested review from arangatang and bidesh October 30, 2025 14:42

arangatang reviewed Oct 30, 2025

View reviewed changes

kyoncal force-pushed the forward-region-via-meta branch from 3795fd7 to b1d59b4 Compare October 30, 2025 16:06

kyoncal requested a review from arangatang October 30, 2025 16:07

arangatang assigned kyoncal Oct 30, 2025

feat(sigv4_helper): inject AWS_REGION in _meta

b9e509f

kyoncal marked this pull request as draft November 3, 2025 10:48

arangatang and others added 2 commits November 3, 2025 17:04

Override the sigv4 signature when adding _meta.

eeee136

feat(sigv4_helper): add region and service argument to _inject_metada…

094fb77

…ta_hook to allow for proper resigning of sigv4 to work

kyoncal force-pushed the forward-region-via-meta branch from b1d59b4 to 094fb77 Compare November 4, 2025 00:55

kyoncal marked this pull request as ready for review November 4, 2025 00:56

arangatang reviewed Nov 4, 2025

View reviewed changes

feat(server.py): add forwarding region as optional argument

6eb9fc0

kyoncal closed this Nov 4, 2025

kyoncal mentioned this pull request Nov 5, 2025

Forward region via meta #71

Draft

8 tasks



		@pytest.mark.asyncio(loop_scope='module')
		async def test_metadata_injection_aws_region(

feat(sigv4_helper): inject AWS_REGION in _meta #62

feat(sigv4_helper): inject AWS_REGION in _meta #62

Uh oh!

Conversation

kyoncal commented Oct 30, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

User experience

Checklist

Acknowledgment

Uh oh!

arangatang left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

kyoncal Oct 30, 2025 • 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!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

kyoncal commented Nov 3, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

kyoncal commented Oct 30, 2025 •

edited

Loading

kyoncal Oct 30, 2025 •

edited

Loading