Implement proper buffering of the output

Signed-off-by: Aivin V. Solatorio <avsolatorio@gmail.com>

mcp_openai_client.py

@@ -65,6 +65,7 @@ When responding you must always plan the steps and enumerate all the tools that
    - Present the data in clear, user-friendly language.
    - You may summarize or explain the data retrieved, but do **not** elaborate based on outside or implicit knowledge.
    - You may describe the data in a way that is easy to understand but you MUST NOT elaborate based on external knowledge.
+   - Provide summary of the answer in the last step describing some observations and insights solely based on the data.
 
 6. **Presentation**:
    - Present the data in a way that is easy to understand.
@@ -215,18 +216,21 @@ class MCPClientWrapper:
                             }
                         )
                     messages[-1]["content"] += partial[-1]["delta"]
-                elif partial[-1].get("response_id"):
-                    previous_response_id = partial[-1]["response_id"]
-                    yield (
-                        messages,
-                        gr.Textbox(value=""),
-                        gr.Textbox(value=previous_response_id),
-                    )
-                    await asyncio.sleep(0.01)
-                    continue
+                    if partial[-1].get("status") == "done":
+                        await asyncio.sleep(0.05)
                 else:
                     is_delta = False
-                    messages.extend(partial)
+                    if partial[-1].get("response_id"):
+                        previous_response_id = partial[-1]["response_id"]
+                        yield (
+                            messages,
+                            gr.Textbox(value=""),
+                            gr.Textbox(value=previous_response_id),
+                        )
+                        await asyncio.sleep(0.01)
+                        continue
+                    else:
+                        messages.extend(partial)
                     print(partial)
 
                 yield (
@@ -378,6 +382,19 @@ class MCPClientWrapper:
                         is_tool_call = False
                         tool_name = None
                         tool_args = None
+
+            elif (
+                isinstance(event, ResponseContentPartDoneEvent)
+                and event.type == "response.content_part.done"
+            ):
+                yield [
+                    {
+                        "role": "assistant",
+                        "content": "",
+                        "delta": "",
+                        "status": "done",
+                    }
+                ]
             elif isinstance(event, ResponseTextDeltaEvent):
                 yield [{"role": "assistant", "content": None, "delta": event.delta}]