[GH-ISSUE #50] Bug: MCP tool results return "(empty result)" due to Pydantic model handling in extract_text_content() #31

New issue

Closed

opened 2026-02-27 07:17:34 +03:00 by kerem · 1 comment

kerem commented

2026-02-27 07:17:34 +03:00

Owner

Originally created by @bhaskoro-muthohar on GitHub (Jan 21, 2026).
Original GitHub issue: https://github.com/jwadow/kiro-gateway/issues/50

Description

MCP tool results are being converted to "(empty result)" instead of preserving the actual content. This affects all MCP servers that return tool results in the standard format [{"type": "text", "text": "..."}].

Root Cause

When Claude Code sends MCP tool results, the content comes as:

{"type": "tool_result", "tool_use_id": "...", "content": [{"type": "text", "text": "actual data here"}]}

Pydantic parses this into:

ToolResultContentBlock(content=[TextContentBlock(type='text', text='actual data here')])

However, extract_text_content() in kiro/converters_core.py only handles dict and str types:

for item in content:
    if isinstance(item, dict):  # False for Pydantic models!
        ...
    elif isinstance(item, str):
        ...

Since TextContentBlock is a Pydantic model (not a dict), the content is not extracted, resulting in an empty string that falls back to "(empty result)".

Steps to Reproduce

Configure any MCP server (e.g., slack-mcp-server, context7)
Use Claude Code with kiro-gateway as the proxy
Call any MCP tool
Observe that the model receives "(empty result)" instead of actual data

Test Case

from kiro.converters_core import extract_text_content
from kiro.models_anthropic import TextContentBlock

# Dict format works
dict_content = [{'type': 'text', 'text': 'Hello'}]
assert extract_text_content(dict_content) == "Hello"  # PASS

# Pydantic format fails
pydantic_content = [TextContentBlock(type='text', text='Hello')]
assert extract_text_content(pydantic_content) == "Hello"  # FAIL - returns ""

Proposed Fix

Add handling for Pydantic models in extract_text_content() (around line 143):

elif hasattr(item, "text"):
    # Handle Pydantic models like TextContentBlock
    text_parts.append(getattr(item, "text", ""))

Impact

All MCP tool results are affected
MCP resources work fine (they return plain strings, not wrapped in content blocks)
This explains why users see data in Claude Code terminal output but the model claims it received empty results

Environment

kiro-gateway version: latest
Python: 3.x with Pydantic v2
Affected MCP servers: slack-mcp-server, context7, likely all others

Originally created by @bhaskoro-muthohar on GitHub (Jan 21, 2026). Original GitHub issue: https://github.com/jwadow/kiro-gateway/issues/50 ## Description MCP tool results are being converted to `"(empty result)"` instead of preserving the actual content. This affects all MCP servers that return tool results in the standard format `[{"type": "text", "text": "..."}]`. ## Root Cause When Claude Code sends MCP tool results, the content comes as: ```json {"type": "tool_result", "tool_use_id": "...", "content": [{"type": "text", "text": "actual data here"}]} ``` Pydantic parses this into: ```python ToolResultContentBlock(content=[TextContentBlock(type='text', text='actual data here')]) ``` However, `extract_text_content()` in `kiro/converters_core.py` only handles `dict` and `str` types: ```python for item in content: if isinstance(item, dict): # False for Pydantic models! ... elif isinstance(item, str): ... ``` Since `TextContentBlock` is a Pydantic model (not a dict), the content is not extracted, resulting in an empty string that falls back to `"(empty result)"`. ## Steps to Reproduce 1. Configure any MCP server (e.g., slack-mcp-server, context7) 2. Use Claude Code with kiro-gateway as the proxy 3. Call any MCP tool 4. Observe that the model receives `"(empty result)"` instead of actual data ## Test Case ```python from kiro.converters_core import extract_text_content from kiro.models_anthropic import TextContentBlock # Dict format works dict_content = [{'type': 'text', 'text': 'Hello'}] assert extract_text_content(dict_content) == "Hello" # PASS # Pydantic format fails pydantic_content = [TextContentBlock(type='text', text='Hello')] assert extract_text_content(pydantic_content) == "Hello" # FAIL - returns "" ``` ## Proposed Fix Add handling for Pydantic models in `extract_text_content()` (around line 143): ```python elif hasattr(item, "text"): # Handle Pydantic models like TextContentBlock text_parts.append(getattr(item, "text", "")) ``` ## Impact - All MCP tool results are affected - MCP resources work fine (they return plain strings, not wrapped in content blocks) - This explains why users see data in Claude Code terminal output but the model claims it received empty results ## Environment - kiro-gateway version: latest - Python: 3.x with Pydantic v2 - Affected MCP servers: slack-mcp-server, context7, likely all others

kerem

2026-02-27 07:17:34 +03:00

closed this issue
added the
bug

fixed
labels

kerem commented

2026-02-27 07:17:34 +03:00

Author

Owner

@jwadow commented on GitHub (Jan 23, 2026):

@bhaskoro-muthohar Your analysis was perfect. Fixed by adding Pydantic model handling to extract_text_content():

elif hasattr(item, "text"):
    text_parts.append(getattr(item, "text", ""))

Added you to CONTRIBUTORS.md

@jwadow commented on GitHub (Jan 23, 2026): @bhaskoro-muthohar Your analysis was perfect. Fixed by adding Pydantic model handling to `extract_text_content()`: ``` elif hasattr(item, "text"): text_parts.append(getattr(item, "text", "")) ``` Added you to CONTRIBUTORS.md