Spaces:
Sleeping
Sleeping
Commit
·
8362096
1
Parent(s):
33f4b8c
Add history
Browse filesSigned-off-by: Aivin V. Solatorio <avsolatorio@gmail.com>
- mcp_openai_client.py +70 -12
mcp_openai_client.py
CHANGED
|
@@ -13,6 +13,7 @@ from anthropic import Anthropic
|
|
| 13 |
from anthropic._exceptions import OverloadedError
|
| 14 |
from dotenv import load_dotenv
|
| 15 |
from openai import OpenAI
|
|
|
|
| 16 |
from openai.types.responses import (
|
| 17 |
ResponseTextDeltaEvent,
|
| 18 |
ResponseContentPartAddedEvent,
|
|
@@ -23,6 +24,7 @@ from openai.types.responses import (
|
|
| 23 |
ResponseMcpCallCompletedEvent,
|
| 24 |
ResponseOutputItemDoneEvent,
|
| 25 |
ResponseOutputItemAddedEvent,
|
|
|
|
| 26 |
)
|
| 27 |
import ast
|
| 28 |
|
|
@@ -171,7 +173,10 @@ class MCPClientWrapper:
|
|
| 171 |
self.session = None
|
| 172 |
|
| 173 |
async def process_message(
|
| 174 |
-
self,
|
|
|
|
|
|
|
|
|
|
| 175 |
):
|
| 176 |
if not self.session and LLM_PROVIDER == "anthropic":
|
| 177 |
messages = history + [
|
|
@@ -181,7 +186,7 @@ class MCPClientWrapper:
|
|
| 181 |
"content": "Please connect to an MCP server first by reloading the page.",
|
| 182 |
},
|
| 183 |
]
|
| 184 |
-
yield messages, gr.Textbox(value="")
|
| 185 |
else:
|
| 186 |
messages = history + [
|
| 187 |
{"role": "user", "content": message},
|
|
@@ -191,13 +196,15 @@ class MCPClientWrapper:
|
|
| 191 |
},
|
| 192 |
]
|
| 193 |
|
| 194 |
-
yield messages, gr.Textbox(value="")
|
| 195 |
# simulate thinking with asyncio.sleep
|
| 196 |
await asyncio.sleep(0.1)
|
| 197 |
messages.pop(-1)
|
| 198 |
|
| 199 |
is_delta = False
|
| 200 |
-
async for partial in self._process_query(
|
|
|
|
|
|
|
| 201 |
if partial[-1].get("delta"):
|
| 202 |
if not is_delta:
|
| 203 |
is_delta = True
|
|
@@ -208,12 +215,25 @@ class MCPClientWrapper:
|
|
| 208 |
}
|
| 209 |
)
|
| 210 |
messages[-1]["content"] += partial[-1]["delta"]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 211 |
else:
|
| 212 |
is_delta = False
|
| 213 |
messages.extend(partial)
|
| 214 |
print(partial)
|
| 215 |
|
| 216 |
-
yield
|
|
|
|
|
|
|
|
|
|
|
|
|
| 217 |
await asyncio.sleep(0.01)
|
| 218 |
|
| 219 |
if (
|
|
@@ -227,7 +247,10 @@ class MCPClientWrapper:
|
|
| 227 |
fl.write(json.dumps(dict(time=f"{datetime.now()}", messages=messages)))
|
| 228 |
|
| 229 |
async def _process_query_openai(
|
| 230 |
-
self,
|
|
|
|
|
|
|
|
|
|
| 231 |
):
|
| 232 |
response = self.openai.responses.create(
|
| 233 |
model=OPENAI_MODEL,
|
|
@@ -247,14 +270,24 @@ class MCPClientWrapper:
|
|
| 247 |
input=message,
|
| 248 |
parallel_tool_calls=False,
|
| 249 |
stream=True,
|
|
|
|
| 250 |
temperature=0,
|
|
|
|
|
|
|
|
|
|
| 251 |
)
|
| 252 |
|
| 253 |
is_tool_call = False
|
| 254 |
tool_name = None
|
| 255 |
tool_args = None
|
| 256 |
for event in response:
|
| 257 |
-
if (
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 258 |
isinstance(event, ResponseOutputItemAddedEvent)
|
| 259 |
and event.item.type == "mcp_call"
|
| 260 |
):
|
|
@@ -553,14 +586,36 @@ class MCPClientWrapper:
|
|
| 553 |
contents.extend(next_response.content)
|
| 554 |
|
| 555 |
async def _process_query(
|
| 556 |
-
self,
|
|
|
|
|
|
|
|
|
|
| 557 |
):
|
| 558 |
if LLM_PROVIDER == "anthropic":
|
| 559 |
async for partial in self._process_query_anthropic(message, history):
|
| 560 |
yield partial
|
| 561 |
elif LLM_PROVIDER == "openai":
|
| 562 |
-
|
| 563 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 564 |
|
| 565 |
|
| 566 |
def gradio_interface(
|
|
@@ -625,6 +680,9 @@ def gradio_interface(
|
|
| 625 |
layout="panel",
|
| 626 |
placeholder="Ask development data questions!",
|
| 627 |
)
|
|
|
|
|
|
|
|
|
|
| 628 |
|
| 629 |
with gr.Row(equal_height=True):
|
| 630 |
msg = gr.Textbox(
|
|
@@ -647,8 +705,8 @@ def gradio_interface(
|
|
| 647 |
|
| 648 |
msg.submit(
|
| 649 |
client.process_message,
|
| 650 |
-
[msg, chatbot],
|
| 651 |
-
[chatbot, msg],
|
| 652 |
concurrency_limit=10,
|
| 653 |
)
|
| 654 |
# clear_btn.click(lambda: [], None, chatbot)
|
|
|
|
| 13 |
from anthropic._exceptions import OverloadedError
|
| 14 |
from dotenv import load_dotenv
|
| 15 |
from openai import OpenAI
|
| 16 |
+
import openai
|
| 17 |
from openai.types.responses import (
|
| 18 |
ResponseTextDeltaEvent,
|
| 19 |
ResponseContentPartAddedEvent,
|
|
|
|
| 24 |
ResponseMcpCallCompletedEvent,
|
| 25 |
ResponseOutputItemDoneEvent,
|
| 26 |
ResponseOutputItemAddedEvent,
|
| 27 |
+
ResponseCompletedEvent,
|
| 28 |
)
|
| 29 |
import ast
|
| 30 |
|
|
|
|
| 173 |
self.session = None
|
| 174 |
|
| 175 |
async def process_message(
|
| 176 |
+
self,
|
| 177 |
+
message: str,
|
| 178 |
+
history: List[Union[Dict[str, Any], ChatMessage]],
|
| 179 |
+
previous_response_id: str = None,
|
| 180 |
):
|
| 181 |
if not self.session and LLM_PROVIDER == "anthropic":
|
| 182 |
messages = history + [
|
|
|
|
| 186 |
"content": "Please connect to an MCP server first by reloading the page.",
|
| 187 |
},
|
| 188 |
]
|
| 189 |
+
yield messages, gr.Textbox(value=""), gr.Textbox(value=previous_response_id)
|
| 190 |
else:
|
| 191 |
messages = history + [
|
| 192 |
{"role": "user", "content": message},
|
|
|
|
| 196 |
},
|
| 197 |
]
|
| 198 |
|
| 199 |
+
yield messages, gr.Textbox(value=""), gr.Textbox(value=previous_response_id)
|
| 200 |
# simulate thinking with asyncio.sleep
|
| 201 |
await asyncio.sleep(0.1)
|
| 202 |
messages.pop(-1)
|
| 203 |
|
| 204 |
is_delta = False
|
| 205 |
+
async for partial in self._process_query(
|
| 206 |
+
message, history, previous_response_id
|
| 207 |
+
):
|
| 208 |
if partial[-1].get("delta"):
|
| 209 |
if not is_delta:
|
| 210 |
is_delta = True
|
|
|
|
| 215 |
}
|
| 216 |
)
|
| 217 |
messages[-1]["content"] += partial[-1]["delta"]
|
| 218 |
+
elif partial[-1].get("response_id"):
|
| 219 |
+
previous_response_id = partial[-1]["response_id"]
|
| 220 |
+
yield (
|
| 221 |
+
messages,
|
| 222 |
+
gr.Textbox(value=""),
|
| 223 |
+
gr.Textbox(value=previous_response_id),
|
| 224 |
+
)
|
| 225 |
+
await asyncio.sleep(0.01)
|
| 226 |
+
continue
|
| 227 |
else:
|
| 228 |
is_delta = False
|
| 229 |
messages.extend(partial)
|
| 230 |
print(partial)
|
| 231 |
|
| 232 |
+
yield (
|
| 233 |
+
messages,
|
| 234 |
+
gr.Textbox(value=""),
|
| 235 |
+
gr.Textbox(value=previous_response_id),
|
| 236 |
+
)
|
| 237 |
await asyncio.sleep(0.01)
|
| 238 |
|
| 239 |
if (
|
|
|
|
| 247 |
fl.write(json.dumps(dict(time=f"{datetime.now()}", messages=messages)))
|
| 248 |
|
| 249 |
async def _process_query_openai(
|
| 250 |
+
self,
|
| 251 |
+
message: str,
|
| 252 |
+
history: List[Union[Dict[str, Any], ChatMessage]],
|
| 253 |
+
previous_response_id: str = None,
|
| 254 |
):
|
| 255 |
response = self.openai.responses.create(
|
| 256 |
model=OPENAI_MODEL,
|
|
|
|
| 270 |
input=message,
|
| 271 |
parallel_tool_calls=False,
|
| 272 |
stream=True,
|
| 273 |
+
max_output_tokens=32768,
|
| 274 |
temperature=0,
|
| 275 |
+
previous_response_id=previous_response_id
|
| 276 |
+
if previous_response_id.strip()
|
| 277 |
+
else None,
|
| 278 |
)
|
| 279 |
|
| 280 |
is_tool_call = False
|
| 281 |
tool_name = None
|
| 282 |
tool_args = None
|
| 283 |
for event in response:
|
| 284 |
+
if isinstance(event, ResponseCompletedEvent):
|
| 285 |
+
yield [
|
| 286 |
+
{
|
| 287 |
+
"response_id": event.response.id,
|
| 288 |
+
}
|
| 289 |
+
]
|
| 290 |
+
elif (
|
| 291 |
isinstance(event, ResponseOutputItemAddedEvent)
|
| 292 |
and event.item.type == "mcp_call"
|
| 293 |
):
|
|
|
|
| 586 |
contents.extend(next_response.content)
|
| 587 |
|
| 588 |
async def _process_query(
|
| 589 |
+
self,
|
| 590 |
+
message: str,
|
| 591 |
+
history: List[Union[Dict[Any, Any], ChatMessage]],
|
| 592 |
+
previous_response_id: str = None,
|
| 593 |
):
|
| 594 |
if LLM_PROVIDER == "anthropic":
|
| 595 |
async for partial in self._process_query_anthropic(message, history):
|
| 596 |
yield partial
|
| 597 |
elif LLM_PROVIDER == "openai":
|
| 598 |
+
try:
|
| 599 |
+
async for partial in self._process_query_openai(
|
| 600 |
+
message, history, previous_response_id
|
| 601 |
+
):
|
| 602 |
+
yield partial
|
| 603 |
+
except openai.APIError as e:
|
| 604 |
+
print(e)
|
| 605 |
+
yield [
|
| 606 |
+
{
|
| 607 |
+
"role": "assistant",
|
| 608 |
+
"content": "The LLM encountered an error. Please try again or reload the page.",
|
| 609 |
+
}
|
| 610 |
+
]
|
| 611 |
+
except Exception as e:
|
| 612 |
+
print(e)
|
| 613 |
+
yield [
|
| 614 |
+
{
|
| 615 |
+
"role": "assistant",
|
| 616 |
+
"content": f"Sorry, I encountered an unexpected error: `{e}`. Please try again or reload the page.",
|
| 617 |
+
}
|
| 618 |
+
]
|
| 619 |
|
| 620 |
|
| 621 |
def gradio_interface(
|
|
|
|
| 680 |
layout="panel",
|
| 681 |
placeholder="Ask development data questions!",
|
| 682 |
)
|
| 683 |
+
previous_response_id = gr.Textbox(
|
| 684 |
+
label="Previous Response ID", interactive=False, visible=False
|
| 685 |
+
)
|
| 686 |
|
| 687 |
with gr.Row(equal_height=True):
|
| 688 |
msg = gr.Textbox(
|
|
|
|
| 705 |
|
| 706 |
msg.submit(
|
| 707 |
client.process_message,
|
| 708 |
+
[msg, chatbot, previous_response_id],
|
| 709 |
+
[chatbot, msg, previous_response_id],
|
| 710 |
concurrency_limit=10,
|
| 711 |
)
|
| 712 |
# clear_btn.click(lambda: [], None, chatbot)
|