fix: preserve MCP embedded resource text

[haha0815]

Summary

• Preserve text from MCP EmbeddedResource.resource.text blocks when rendering tool results.
• Add regression coverage for embedded resource text in structured MCP output.

Tests

• ./venv/bin/python3 -m pytest tests/tools/test_mcp_structured_content.py tests/tools/test_mcp_image_content.py tests/tools/test_mcp_tool_session_expired.py -q → 34 passed

Notes

• Fix was verified in a fresh Hermes process for QMD get returning the expected document text instead of an empty payload.

[alt-glitch]

Duplicate of open PR #13736 which implements the same EmbeddedResource text extraction fix. Please coordinate with that PR.

[teknium1]

Thanks for the focused regression coverage. The premise is confirmed on current main: tools/mcp_tool.py:3950 and tools/mcp_tool.py:3971 only consume direct block.text, so EmbeddedResource.resource.text is currently discarded. The proposed helper correctly applies the text extraction to both the normal and isError paths.

Suggested changes

• If this is salvaged, cover EmbeddedResource.resource.blob as well as text. The related MCP fix discussion for fix(mcp): extract text from EmbeddedResource content blocks in tool results #13736/fix(mcp): preserve embedded resource text in tool results (#30601) #31356 treats blobs as the sibling resource shape, and tools/mcp_tool.py:4129 already renders binary read-resource payloads as a placeholder.

This is an automated hermes-sweeper review.

[teknium1]

This bug is now fixed on main via PR #64556 (salvaged from #64061). Your PR was the earliest report and fix for the dropped EmbeddedResource text — submitted May 21, well before the others — and that priority is acknowledged. We merged the #64061 implementation because it additionally materializes binary blob resources (PDFs etc.) to the document cache, handles ResourceLink and Audio blocks, and ships size caps + hostile-input tests. Thanks for catching this first!