direct_proxy resource read raises AttributeError ('…ResourceContents' has no attribute 'id') and returns empty content

[jedwards1230]

🐞 Bug Summary

With MCPGATEWAY_DIRECT_PROXY_ENABLED=true and a gateway in gateway_mode=direct_proxy, a resources/read from a normal MCP client returns empty content. The direct_proxy branch successfully fetches the resource from the upstream and builds a TextResourceContents/BlobResourceContents, then the shared post-fetch block calls getattr(content, "id") on that object — which has no id field — and raises AttributeError. The exception is caught at the transport layer and turned into empty content, so the client never sees the fetched data.

This affects the path a normal client takes (no X-Context-Forge-Gateway-Id header). With that header set, the request uses a different transport-level proxy path that returns the content correctly.

---

🧩 Affected Component

• Federation or Transports

---

🔁 Steps to Reproduce

1. Set MCPGATEWAY_DIRECT_PROXY_ENABLED=true and restart the gateway.
2. Federate an upstream MCP server that exposes a resource, then set the gateway to direct_proxy: PUT /gateways/{id} with {"gateway_mode":"direct_proxy"} (the gateway also needs transport: STREAMABLEHTTP).
3. From an MCP client, resources/read the federated resource through /servers/{id}/mcp — without sending the X-Context-Forge-Gateway-Id header.

Reproduced locally against a FastMCP upstream federated into a local gateway (sqlite), for both a text and a binary (image/png) resource.

---

🤔 Expected Behavior

The read returns the upstream resource's content (text or blob), fetched live. Instead the client receives empty content:

ReadResourceResult(contents=[TextResourceContents(uri=AnyUrl('test://schema'), mimeType='text/plain', text='')])

---

📓 Logs / Error Output

ERROR mcpgateway.transports.streamablehttp_transport - Error reading resource 'test://schema': 'TextResourceContents' object has no attribute 'id'
  File ".../mcpgateway/services/resource_service.py", line 2549, in read_resource
    resource_id=getattr(content, "id"),
AttributeError: 'TextResourceContents' object has no attribute 'id'

The binary resource fails identically with 'BlobResourceContents' object has no attribute 'id' at the same line.

---

🧠 Environment Info

Key	Value
Version or commit	v1.0.4 (also present on main)
Runtime	Python 3.12.13
Platform / OS	Linux / macOS (reproduced locally, sqlite)
Container	mcp-contextforge-gateway==1.0.4
Flags	MCPGATEWAY_DIRECT_PROXY_ENABLED=true, gateway gateway_mode=direct_proxy

---

🧩 Additional Context

The direct_proxy branch in resource_service.py::read_resource (around lines 2381–2432, gated on gateway.gateway_mode == "direct_proxy" and settings.mcpgateway_direct_proxy_enabled) fetches over streamablehttp_client and constructs a TextResourceContents or BlobResourceContents from mcpgateway/common/models.py. Those classes define only uri, mime_type, meta, and text/blob — there is no id field.

Because both subclass ResourceContents, execution then falls into the post-fetch "RESOLVE CONTENT" block (around lines 2541–2576). Its first branch, if isinstance(content, (ResourceContent, ResourceContents, TextContent)): (line 2545), matches, and the next statement runs resource_id=getattr(content, "id") (line 2549) with no default, raising AttributeError. There is no early return between the direct_proxy fetch and that call, so every direct_proxy text or blob read hits it. The handler in streamablehttp_transport.py catches Exception and returns "".

The test suite works around this by monkey-patching id-bearing subclasses over the real models (tests/unit/mcpgateway/services/test_resource_service.py ~7127), with a comment noting the post-direct-proxy code calls getattr(content, "id") unconditionally — so the production classes that lack id take a path the tests never exercise.

A fix would return the constructed TextResourceContents/BlobResourceContents from the direct_proxy branch directly rather than routing it through the legacy getattr(content, "id") post-processing, or guard that block with getattr(content, "id", None) plus a blob-aware path. A test asserting that a direct_proxy text read and an image/png read both return content end-to-end would cover it.

[Ewertonslv]

I hit this too and traced it. In direct_proxy mode, ResourceService.read_resource builds the final TextResourceContents / BlobResourceContents and then falls through into the shared post-fetch RESOLVE CONTENT block, whose first branch calls getattr(content, "id") unconditionally. The MCP-compliant models in mcpgateway/common/models.py have no id field, so every direct_proxy text/blob read raises AttributeError; the transport swallows it and the client silently gets text="". The # Skip the rest of the DB lookup logic comment there never actually skipped anything.

I have a fix with regression tests up at #5463 (a direct_proxy_read guard added as the first branch of that if/elif, ~12 lines in resource_service.py). The existing direct_proxy tests masked the bug by monkey-patching id-bearing subclasses over the real models, so the added tests exercise the production classes.

This issue is still labeled triage, and per CONTRIBUTING maintainers won't review a PR while that's the case — so it's blocked on triage rather than on code. Could a maintainer confirm the diagnosis and drop the triage label if the fix is wanted? Happy to adjust the approach if you'd prefer the skip handled differently.