fix(mcp): extract text from EmbeddedResource content blocks in tool results

[jpterry]

What does this PR do?

MCP tools that return file/document content use EmbeddedResource content blocks (type: "resource") where the text payload lives at block.resource.text (per MCP spec §5.6.2 — Tool Result Content). The tool result handler in _make_tool_handler only checked for block.text (which exists on TextContent blocks), silently dropping EmbeddedResource blocks and returning empty strings to the agent.

Discovered via the @tobilu/qmd MCP server (v2.1.0), whose get and multi_get tools return EmbeddedResource blocks for document retrieval. Any MCP server returning resource-type content blocks would hit the same issue.

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)

Changes Made

• tools/mcp_tool.py: Added elif hasattr(block, "resource") branch in _make_tool_handler's content extraction loop to handle block.resource.text (TextResourceContents) and block.resource.blob (BlobResourceContents)
• tests/tools/test_mcp_tool.py: Added three tests:
  • test_embedded_resource_text — EmbeddedResource with text content
  • test_embedded_resource_blob — binary blob produces placeholder
  • test_mixed_text_and_embedded_resource — mixed TextContent + EmbeddedResource

How to Test

1. Configure any MCP server that returns EmbeddedResource content blocks (e.g. @tobilu/qmd v2.1.0 with get tool)
2. Call a tool that retrieves document content (e.g. mcp_qmd_get)
3. Before fix: Tool executes successfully but returns empty string
4. After fix: Document content is correctly extracted and returned

Or run the unit tests: pytest tests/tools/test_mcp_tool.py -q (169 pass, 3 new)

Checklist

Code

• I've read the Contributing Guide
• Commit messages follow Conventional Commits
• Searched for existing PRs — no duplicate
• PR contains only changes related to this fix
• pytest tests/ -q passes (169 MCP tool tests)
• Added 3 tests for the fix
• Tested on macOS 26.4 (Apple Silicon)

Documentation & Housekeeping

• N/A — no config or documentation changes needed
• Cross-platform: change is pure Python, no platform-specific code

[alt-glitch]

Related to #3655 (closed earlier attempt at same fix) and #10848 (similar pattern for ImageContent blocks). All stem from incomplete content-type handling in _make_tool_handler.

[alt-glitch]

Related to #3655 and #10848.

[nadunperera]

I hit the same issue via QMD get / multi_get and accidentally opened duplicate #18118 before finding this PR. I’m closing mine in favor of this one.

One small implementation note from the duplicate: I routed both normal and isError result content through the same extraction helper, so EmbeddedResource text is not dropped in error payloads either. Not required for the QMD happy path, but may be worth folding in if you want the handling to be consistent.

[jpterry]

Is there anything I can do to move this along?
I'd be happy to resolve conflicts and/or even pull in related changes to this PR.
This is my first PR here, so let me know if I've missed some steps.

[jpterry]

Updated branch to address conflicts. Can this PR be approved to run the CI checks?

[jpterry]

#31356 looks like a more complete fix than this one.