fix: crash on non-text content blocks in Anthropic LLM

[Giggitycountless]

What changed

Fix 4 crash sites in augmented_llm_anthropic.py where block.text is accessed unconditionally on non-text content blocks (ThinkingBlock, RedactedThinkingBlock, ToolUseBlock).

Bug

The Anthropic API returns Message.content as a union of TextBlock | ToolUseBlock | ThinkingBlock | RedactedThinkingBlock. Only TextBlock has a .text attribute. Three functions accessed .text unconditionally:

Function	Line	Error
message_param_str	915	AttributeError: dict has no attribute text
message_str	932	AttributeError: ThinkingBlock has no attribute text
anthropic_content_to_mcp_content	1378	AttributeError: ToolUseBlock has no attribute text
anthropic_content_to_mcp_content	1381	KeyError: text

This crashes whenever extended thinking or tool use is active in the response.

Fix

• message_param_str and message_str: Add hasattr(block, "text") guard before accessing .text
• anthropic_content_to_mcp_content: Use getattr(block, "text", None) for pydantic models and block.get("text") for dicts

4 lines changed total.

Tests

10 new test cases covering all 4 crash sites:

• TextBlock (baseline, should work)
• ToolUseBlock (pydantic model, no .text)
• ThinkingBlock (pydantic model, no .text)
• RedactedThinkingBlock (pydantic model, no .text)
• Mixed blocks (text + tool_use + thinking + redacted)
• Dict-style tool_use (no "text" key)
• Dict-style thinking (no "text" key)
• message_str with ThinkingBlock
• message_str with ToolUseBlock
• message_param_str with non-text dict blocks

All 36 tests pass (26 existing + 10 new). Zero regressions.

Summary by CodeRabbit

Release Notes

• Bug Fixes

  • Enhanced stability when processing content blocks with missing or non-standard attributes, preventing crashes in edge cases.
  • Improved error handling for message formatting with unexpected content types.

• Tests

  • Added test coverage for edge cases involving various content block types and missing expected attributes.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: cd155cf9-54b3-40b4-9aaa-2d06ca0868d9

📥 Commits

Reviewing files that changed from the base of the PR and between f62d849 and b7567d4.

📒 Files selected for processing (2)

• src/mcp_agent/workflows/llm/augmented_llm_anthropic.py
• tests/workflows/llm/test_augmented_llm_anthropic.py

---

📝 Walkthrough

Walkthrough

This change hardens the Anthropic-to-MCP message conversion logic to safely handle content blocks that may lack a .text attribute or be non-dict objects. Safe attribute extraction replaces required access patterns, with fallback to None or direct stringification when text is unavailable.

Changes

Cohort / File(s)	Summary
Message Safety Hardening src/mcp_agent/workflows/llm/augmented_llm_anthropic.py	Updated message_param_str, message_str, and anthropic_content_to_mcp_content to use safe attribute/key access (getattr(), .get()) instead of required access patterns when extracting block text, preventing errors on non-text block types.
Test Coverage for Non-Text Blocks tests/workflows/llm/test_augmented_llm_anthropic.py	Added comprehensive test suite validating that anthropic_content_to_mcp_content, AnthropicAugmentedLLM.message_str, and message_param_str handle non-text Anthropic content blocks (thinking, redacted_thinking, tool_use) without raising attribute or key errors.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Suggested reviewers

• saqadri

Poem

🐰 A rabbit hops through blocks with care,
Safe attributes everywhere!
No crashes when .text is gone,
Your code stands firm and carries on! ✨

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately describes the main change: fixing crashes when handling non-text content blocks in Anthropic LLM integration.
Docstring Coverage	✅ Passed	Docstring coverage is 92.86% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Review rate limit: 7/8 reviews remaining, refill in 7 minutes and 30 seconds.}

_{Comment @coderabbitai help to get the list of available commands and usage tips.}