Add test for mcp result text extraction

Author: Mofef

integrations/mcp/src/haystack_integrations/tools/mcp/mcp_tool.py

@@ -63,7 +63,7 @@ def _resolve_headers(headers: dict[str, str | Secret] | None) -> dict[str, str]
     return resolved_headers
 
 
-def extract_first_text(result: str) -> str:
+def extract_first_text_element(result: str) -> str:
     # Per MCP spec, content[] may contain TextContent, ImageContent, AudioContent, etc.
     # Parse only first TextContent block (ToolInvoker requires dict, not list).
     parsed: dict = json.loads(result)
@@ -1103,7 +1103,7 @@ async def invoke() -> Any:
             # Parse JSON to dict only when outputs_to_state is configured.
             # ToolInvoker requires dict for _merge_tool_outputs(); ToolCallResult.result expects str otherwise.
             if self.outputs_to_state:
-                return self.extract_first_text(result)
+                return extract_first_text_element(result)
 
             return result
         except (MCPError, TimeoutError) as e:
@@ -1134,7 +1134,7 @@ async def ainvoke(self, **kwargs: Any) -> str | dict[str, Any]:
             # Parse JSON to dict only when outputs_to_state is configured.
             # ToolInvoker requires dict for _merge_tool_outputs(); ToolCallResult.result expects str otherwise.
             if self.outputs_to_state:
-                return extract_first_text(result)
+                return extract_first_text_element(result)
 
             return result
         except asyncio.TimeoutError as e:

integrations/mcp/tests/test_mcp_tool.py

@@ -15,7 +15,7 @@
     MCPTool,
     StdioServerInfo,
 )
-from haystack_integrations.tools.mcp.mcp_tool import StdioClient
+from haystack_integrations.tools.mcp.mcp_tool import StdioClient, extract_first_text_element
 
 from .mcp_memory_transport import InMemoryServerInfo
 from .mcp_servers_fixtures import calculator_mcp, echo_mcp
@@ -27,6 +27,23 @@ def simple_haystack_tool(name: str) -> str:
     return f"Hello, {name}!"
 
 
+def test_extract_first_text_element():
+    """Test that extract_first_text skips non-text blocks and parses the first text block."""
+    result = json.dumps(
+        {
+            "content": [
+                {"type": "image", "data": "ignored"},
+                {"type": "text", "text": '{"answer": 42}'},
+                {"type": "text", "text": '{"answer": 7}'},
+            ]
+        }
+    )
+
+    extracted = extract_first_text_element(result)
+
+    assert extracted == {"answer": 42}
+
+
 class TestMCPTool:
     """Tests for the MCPTool class using in-memory servers."""