client2 / app.py
PD03's picture
Update app.py
77f67f6 verified
raw
history blame
19.5 kB
import gradio as gr
import openai
import requests
import json
from typing import Dict, Any, List, Tuple
from datetime import datetime
import os
class MCPClient:
"""MCP Client for communicating with the MCP server"""
def __init__(self, server_url: str):
self.server_url = server_url.rstrip('/')
def call_tool_sync(self, tool_name: str, arguments: Dict[str, Any] = None) -> Dict[str, Any]:
"""Synchronous tool call using requests instead of aiohttp"""
if arguments is None:
arguments = {}
mcp_request = {
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": {
"name": tool_name,
"arguments": arguments
}
}
try:
response = requests.post(
f"{self.server_url}/mcp",
json=mcp_request,
headers={
"Content-Type": "application/json",
"ngrok-skip-browser-warning": "true"
},
timeout=30
)
if response.status_code == 200:
result = response.json()
if "result" in result and "content" in result["result"]:
content = result["result"]["content"][0]["text"]
return json.loads(content)
return result
else:
return {
"success": False,
"error": f"HTTP {response.status_code}: {response.text}"
}
except Exception as e:
return {
"success": False,
"error": f"Connection error: {str(e)}"
}
def list_tools_sync(self) -> List[Dict[str, Any]]:
"""Synchronous tool listing using requests"""
mcp_request = {
"jsonrpc": "2.0",
"id": 1,
"method": "tools/list"
}
try:
response = requests.post(
f"{self.server_url}/mcp",
json=mcp_request,
headers={
"Content-Type": "application/json",
"ngrok-skip-browser-warning": "true"
},
timeout=30
)
if response.status_code == 200:
result = response.json()
return result.get("result", {}).get("tools", [])
return []
except Exception as e:
print(f"Error listing tools: {str(e)}")
return []
class AIAssistant:
"""AI Assistant with MCP integration"""
def __init__(self, openai_api_key: str, mcp_client: MCPClient):
try:
self.openai_client = openai.OpenAI(
api_key=openai_api_key,
timeout=30.0
)
except Exception as e:
# Fallback for older OpenAI versions
openai.api_key = openai_api_key
self.openai_client = openai
self.mcp_client = mcp_client
self.available_tools = []
def initialize(self):
"""Initialize the assistant by fetching available tools"""
self.available_tools = self.mcp_client.list_tools_sync()
def get_system_prompt(self) -> str:
"""Generate system prompt with available tools"""
tools_description = "\n".join([
f"- {tool['name']}: {tool['description']}"
for tool in self.available_tools
])
return f"""You are an AI assistant with access to SAP business systems and news data through specialized tools.
Available tools:
{tools_description}
When a user asks for information that can be retrieved using these tools, you should:
1. Identify which tool(s) would be helpful
2. Call the appropriate tool(s) with the right parameters
3. Wait for the results before providing your final response
For SAP-related queries (purchase orders, requisitions), use the SAP tools.
For news-related queries, use the news tools.
To call a tool, use this exact format:
CALL_TOOL: tool_name
or
CALL_TOOL: tool_name(parameter1=value1, parameter2=value2)
Examples:
- For "show me purchase orders": CALL_TOOL: get_purchase_orders
- For "get 20 purchase orders": CALL_TOOL: get_purchase_orders(top=20)
- For "latest tech news": CALL_TOOL: get_news_headlines(category=technology)
- For "get news from BBC": CALL_TOOL: get_news_by_source(source_id=bbc-news)
- For "get news from CNN": CALL_TOOL: get_news_by_source(source_id=cnn)
- For "get news from Reuters": CALL_TOOL: get_news_by_source(source_id=reuters)
IMPORTANT: For news by source queries, always include the source_id parameter:
- BBC: source_id=bbc-news
- CNN: source_id=cnn
- Reuters: source_id=reuters
- Associated Press: source_id=associated-press
- The Guardian: source_id=the-guardian
- Washington Post: source_id=the-washington-post
After calling a tool, I will provide you with the results to interpret for the user.
"""
def extract_tool_calls(self, response: str) -> List[Dict[str, Any]]:
"""Extract tool calls from AI response"""
tool_calls = []
lines = response.split('\n')
for line in lines:
line = line.strip()
if line.startswith('CALL_TOOL:'):
try:
# Remove 'CALL_TOOL:' prefix and clean up
tool_part = line[10:].strip()
# Handle cases with or without parentheses
if '(' in tool_part and ')' in tool_part:
tool_name = tool_part.split('(')[0].strip()
params_str = tool_part.split('(')[1].split(')')[0]
params = {}
if params_str.strip():
for param in params_str.split(','):
if '=' in param:
key, value = param.split('=', 1)
key = key.strip()
value = value.strip().strip('"\'')
try:
if value.isdigit():
value = int(value)
elif value.lower() in ['true', 'false']:
value = value.lower() == 'true'
except:
pass
params[key] = value
tool_calls.append({
'name': tool_name,
'arguments': params
})
else:
# Simple tool call without parameters
tool_name = tool_part.strip()
tool_calls.append({
'name': tool_name,
'arguments': {}
})
except Exception as e:
print(f"Error parsing tool call '{line}': {e}")
continue
return tool_calls
def truncate_tool_result(self, result: Dict[str, Any], max_chars: int = 2000) -> Dict[str, Any]:
"""Truncate tool results to prevent context overflow"""
if not isinstance(result, dict):
return result
result_copy = result.copy()
result_str = json.dumps(result_copy, indent=2)
if len(result_str) > max_chars:
# Try to truncate data arrays/lists first
for key, value in result_copy.items():
if isinstance(value, list) and len(value) > 3:
result_copy[key] = value[:3] + [f"... ({len(value) - 3} more items truncated)"]
elif isinstance(value, str) and len(value) > 500:
result_copy[key] = value[:500] + "... (truncated)"
# If still too long, add truncation notice
result_str = json.dumps(result_copy, indent=2)
if len(result_str) > max_chars:
result_copy = {
"success": result.get("success", False),
"truncated": True,
"message": f"Result truncated due to size. Original had {len(result_str)} characters.",
"sample_data": str(result)[:1000] + "..." if len(str(result)) > 1000 else str(result)
}
return result_copy
def process_message(self, user_message: str) -> Tuple[str, str]:
"""Process user message and handle tool calls"""
tool_info = ""
try:
messages = [
{"role": "system", "content": self.get_system_prompt()},
{"role": "user", "content": user_message}
]
# Check if we have a proper OpenAI client
if hasattr(self.openai_client, 'chat'):
response = self.openai_client.chat.completions.create(
model="gpt-3.5-turbo",
messages=messages,
temperature=0.7,
max_tokens=800 # Reduced to leave more room for context
)
ai_response = response.choices[0].message.content
else:
# Fallback for older API
response = self.openai_client.ChatCompletion.create(
model="gpt-3.5-turbo",
messages=messages,
temperature=0.7,
max_tokens=800
)
ai_response = response.choices[0].message.content
tool_calls = self.extract_tool_calls(ai_response)
# Debug information
print(f"AI Response: {ai_response}")
print(f"Extracted tool calls: {tool_calls}")
if tool_calls:
tool_results = []
for tool_call in tool_calls:
tool_info += f"πŸ”§ Calling: {tool_call['name']}\n"
# FIXED: Use call_tool_sync instead of await call_tool
result = self.mcp_client.call_tool_sync(
tool_call['name'],
tool_call['arguments']
)
# Truncate large results to prevent context overflow
truncated_result = self.truncate_tool_result(result)
tool_results.append({
'tool': tool_call['name'],
'result': truncated_result
})
if result.get('success'):
tool_info += f"βœ… {tool_call['name']} completed\n"
else:
tool_info += f"❌ {tool_call['name']} failed: {result.get('error', 'Unknown error')}\n"
# Create concise tool results summary
tool_results_text = "\n\n".join([
f"Tool: {tr['tool']}\nResult: {json.dumps(tr['result'], indent=2)[:1500]}{'...(truncated)' if len(json.dumps(tr['result'], indent=2)) > 1500 else ''}"
for tr in tool_results
])
final_messages = messages + [
{"role": "assistant", "content": ai_response},
{"role": "user", "content": f"Here are the tool results:\n\n{tool_results_text}\n\nPlease interpret these results and provide a helpful response to the user."}
]
# Get final response with tool results
if hasattr(self.openai_client, 'chat'):
final_response = self.openai_client.chat.completions.create(
model="gpt-3.5-turbo",
messages=final_messages,
temperature=0.7,
max_tokens=800 # Reduced max tokens
)
return final_response.choices[0].message.content, tool_info
else:
final_response = self.openai_client.ChatCompletion.create(
model="gpt-3.5-turbo",
messages=final_messages,
temperature=0.7,
max_tokens=800
)
return final_response.choices[0].message.content, tool_info
else:
return ai_response, ""
except Exception as e:
return f"❌ Error processing your request: {str(e)}", ""
# Global variables
assistant = None
mcp_client = None
def test_connection(mcp_url):
"""Test MCP server connection"""
if not mcp_url or mcp_url == "https://your-ngrok-url.ngrok.io":
return "❌ Please enter a valid MCP server URL"
try:
# Test health endpoint
response = requests.get(f"{mcp_url.rstrip('/')}/health", timeout=10)
if response.status_code == 200:
data = response.json()
# Test MCP tools list
mcp_request = {
"jsonrpc": "2.0",
"id": 1,
"method": "tools/list"
}
mcp_response = requests.post(
f"{mcp_url.rstrip('/')}/mcp",
json=mcp_request,
headers={
"Content-Type": "application/json",
"ngrok-skip-browser-warning": "true"
},
timeout=10
)
if mcp_response.status_code == 200:
mcp_data = mcp_response.json()
tools = mcp_data.get("result", {}).get("tools", [])
tool_names = [tool.get("name", "Unknown") for tool in tools]
return f"βœ… Connected successfully!\nHealth Status: {data.get('status', 'Unknown')}\nMCP Tools: {len(tools)}\nAvailable: {', '.join(tool_names)}"
else:
return f"βœ… Health OK, but MCP endpoint failed: HTTP {mcp_response.status_code}"
else:
return f"❌ Connection failed: HTTP {response.status_code}"
except Exception as e:
return f"❌ Connection error: {str(e)}"
def initialize_assistant(openai_key, mcp_url):
"""Initialize the AI assistant"""
global assistant, mcp_client
if not openai_key:
return "❌ Please enter your OpenAI API key"
if not mcp_url or mcp_url == "https://your-ngrok-url.ngrok.io":
return "❌ Please enter a valid MCP server URL"
try:
mcp_client = MCPClient(mcp_url)
assistant = AIAssistant(openai_key, mcp_client)
assistant.initialize()
return f"βœ… AI Assistant initialized with {len(assistant.available_tools)} tools available"
except Exception as e:
return f"❌ Failed to initialize: {str(e)}"
def chat_interface(message, history, openai_key, mcp_url):
"""Main chat interface"""
global assistant
if not assistant:
init_result = initialize_assistant(openai_key, mcp_url)
if "❌" in init_result:
history.append([message, init_result])
return history, ""
try:
print(f"Calling process_message with: {message}")
# Limit conversation history to prevent context overflow
# Keep only the last 5 exchanges (10 messages total)
if len(history) > 10:
history = history[-10:]
# Make sure we call the synchronous method
result = assistant.process_message(message)
print(f"process_message returned: {type(result)} - {result}")
# Check if result is a tuple (response, tool_info)
if isinstance(result, tuple) and len(result) == 2:
response, tool_info = result
print(f"Unpacked: response={response}, tool_info={tool_info}")
else:
response = str(result)
tool_info = ""
print(f"Single result: {response}")
# Format response with tool info if available
if tool_info:
full_response = f"**Tool Execution:**\n{tool_info}\n\n**Response:**\n{response}"
else:
full_response = response
history.append([message, full_response])
return history, ""
except Exception as e:
import traceback
error_response = f"❌ Error: {str(e)}\n\nTraceback:\n{traceback.format_exc()}"
print(f"Error in chat_interface: {error_response}")
history.append([message, error_response])
return history, ""
# Create Gradio interface
with gr.Blocks(title="AI Assistant with SAP & News Integration", theme=gr.themes.Soft()) as demo:
gr.Markdown("# πŸ€– AI Assistant with SAP & News Integration")
gr.Markdown("Chat with an AI that can access SAP business data and news through natural language queries.")
with gr.Row():
with gr.Column(scale=2):
chatbot = gr.Chatbot(
height=500,
show_label=False,
container=True,
bubble_full_width=False
)
msg = gr.Textbox(
placeholder="Ask me about SAP data, news, or anything else...",
show_label=False,
container=False
)
with gr.Row():
submit_btn = gr.Button("Send", variant="primary")
clear_btn = gr.Button("Clear", variant="secondary")
with gr.Column(scale=1):
gr.Markdown("### βš™οΈ Configuration")
openai_key = gr.Textbox(
label="OpenAI API Key",
type="password",
placeholder="sk-..."
)
mcp_url = gr.Textbox(
label="MCP Server URL",
value="https://your-ngrok-url.ngrok.io",
placeholder="https://abc123.ngrok.io"
)
test_btn = gr.Button("Test Connection", variant="secondary")
connection_status = gr.Textbox(label="Connection Status", interactive=False)
gr.Markdown("### πŸ“‹ Example Queries")
gr.Markdown("""
- "Show me recent purchase orders"
- "Get purchase requisitions"
- "What's the latest tech news?"
- "Get news from BBC"
- "Show me business news from the US"
""")
# Event handlers
def respond(message, history, openai_key, mcp_url):
return chat_interface(message, history, openai_key, mcp_url)
submit_btn.click(
respond,
[msg, chatbot, openai_key, mcp_url],
[chatbot, msg]
)
msg.submit(
respond,
[msg, chatbot, openai_key, mcp_url],
[chatbot, msg]
)
clear_btn.click(lambda: ([], ""), outputs=[chatbot, msg])
test_btn.click(
test_connection,
[mcp_url],
[connection_status]
)
if __name__ == "__main__":
demo.launch()