fix: convert tool result content to string for provider compatibility

[Giggitycountless]

Description

OpenAI's API accepts both string and list[ChatCompletionContentPartTextParam] formats for tool message content, but third-party providers (DeepSeek, etc.) only accept a plain string. This causes deserialization errors:

Failed to deserialize the JSON body into the target type: messages[N]: invalid type: sequence, expected a string

Root Cause

In execute_tool_call(), tool result content was built as a list:

content=[mcp_content_to_openai_content_part(c) for c in result.content]

This creates [{"type": "text", "text": "..."}] which OpenAI accepts but DeepSeek rejects.

Fix

Add _tool_result_to_string() helper that extracts text from MCP content parts and joins them into a plain string. For non-text parts (images, binary resources), includes a descriptive placeholder.

# Before
content=[mcp_content_to_openai_content_part(c) for c in result.content]

# After  
content=self._tool_result_to_string(result.content)

The helper handles all MCP content types:

• TextContent → extracts .text
• EmbeddedResource with TextResourceContents → extracts .resource.text
• EmbeddedResource with BlobResourceContents → [mimeType] placeholder
• ImageContent → [image: mimeType] placeholder

Fixes #25

How Has This Been Tested?

• Verified _tool_result_to_string() correctly handles text, image, and empty content
• Confirmed existing tests still pass (pre-existing generate_stream abstract method error is unrelated)

Checklist

• I have read the Contributing Guide
• My changes follow the established code style
• I have tested my changes locally

Summary by CodeRabbit

Release Notes

• Refactor
  • Improved tool result message formatting for language model integration, now using plain string format for more consistent handling of tool outputs and embedded content.

[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: d9e8fc08-56b0-4e15-9e30-c715a9250716

📥 Commits

Reviewing files that changed from the base of the PR and between f62d849 and ad562da.

📒 Files selected for processing (1)

• src/mcp_agent/workflows/llm/augmented_llm_openai.py

---

📝 Walkthrough

Walkthrough

Tool-call message construction in OpenAI augmented LLM now converts MCP result content to a plain string instead of a list. A new helper method processes TextContent, ImageContent, and EmbeddedResource parts, concatenating text and inserting placeholders for non-text items.

Changes

Cohort / File(s)	Summary
Tool Result Content Formatting src/mcp_agent/workflows/llm/augmented_llm_openai.py	Added static method _tool_result_to_string() that transforms MCP content parts into a single concatenated string for OpenAI ChatCompletionToolMessageParam, handling TextContent, ImageContent, and EmbeddedResource objects with appropriate placeholders for non-text items.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

• Fix empty mcp tool result error #330: Also modifies tool-call handling for empty MCP tool results in the same file, adding logic to handle ChatCompletionToolMessageParam construction—directly related to this PR's content formatting fix.
• Fix openai workflow for handling images #341: Adjusts tool-call content handling in augmented_llm_openai.py for image content formatting and mixed-message returns—overlaps with this PR's content transformation logic.

Suggested reviewers

• rholinshead
• saqadri

Poem

🐰 A rabbit's ode to string conversion:

Lists were causing trouble, causing pain,
OpenAI wanted strings, oh so plain!
Mix the content parts with care and glow,
Text concatenated, steady and slow. ✨
No more list confusion, all is right!

🚥 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: converting tool result content to a string for provider compatibility, which is the core fix in this PR.
Linked Issues check	✅ Passed	The PR successfully addresses issue #25 by converting tool result content from a list to a plain string, preventing deserialization errors in non-OpenAI providers while maintaining compatibility.
Out of Scope Changes check	✅ Passed	All changes are directly related to fixing the tool result content type issue. The addition of the _tool_result_to_string helper and its usage in execute_tool_call are within scope.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.

_{✏️ 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.}