Puch AI MCP Starter

import asyncio from typing import Annotated import os from dotenv import load_dotenv from fastmcp import FastMCP from fastmcp.server.auth.providers.bearer import BearerAuthProvider, RSAKeyPair from mcp import ErrorData, McpError from mcp.server.auth.provider import AccessToken from mcp.types import TextContent, ImageContent, INVALID_PARAMS, INTERNAL_ERROR from pydantic import BaseModel, Field, AnyUrl import markdownify import httpx import readabilipy import json import logging # --- Logging Configuration --- logging.basicConfig(level=logging.INFO) logger = logging.getLogger(__name__) # --- Load environment variables --- load_dotenv() TOKEN = os.environ.get("AUTH_TOKEN") MY_NUMBER = os.environ.get("MY_NUMBER") assert TOKEN is not None, "Please set AUTH_TOKEN in your .env file" assert MY_NUMBER is not None, "Please set MY_NUMBER in your .env file" # --- Auth Provider --- class SimpleBearerAuthProvider(BearerAuthProvider): def __init__(self, token: str): k = RSAKeyPair.generate() super().__init__(public_key=k.public_key, jwks_uri=None, issuer=None, audience=None) self.token = token async def load_access_token(self, token: str) -> AccessToken | None: if token == self.token: return AccessToken( token=token, client_id="puch-client", scopes=["*"], expires_at=None, ) return None # --- Rich Tool Description model --- class RichToolDescription(BaseModel): description: str use_when: str side_effects: str | None = None # --- Fetch Utility Class --- class Fetch: USER_AGENT = "Puch/1.0 (Autonomous)" @classmethod async def fetch_url( cls, url: str, user_agent: str, force_raw: bool = False, ) -> tuple[str, str]: async with httpx.AsyncClient() as client: try: response = await client.get( url, follow_redirects=True, headers={"User-Agent": user_agent}, timeout=30, ) except httpx.HTTPError as e: raise McpError(ErrorData(code=INTERNAL_ERROR, message=f"Failed to fetch {url}: {e!r}")) if response.status_code >= 400: raise McpError(ErrorData(code=INTERNAL_ERROR, message=f"Failed to fetch {url} - status code {response.status_code}")) page_raw = response.text content_type = response.headers.get("content-type", "") is_page_html = "text/html" in content_type if is_page_html and not force_raw: return cls.extract_content_from_html(page_raw), "" return ( page_raw, f"Content type {content_type} cannot be simplified to markdown, but here is the raw content:\n", ) @staticmethod def extract_content_from_html(html: str) -> str: """Extract and convert HTML content to Markdown format.""" ret = readabilipy.simple_json.simple_json_from_html_string(html, use_readability=True) if not ret or not ret.get("content"): return "Page failed to be simplified from HTML" content = markdownify.markdownify(ret["content"], heading_style=markdownify.ATX) return content # --- MCP Server Setup --- mcp = FastMCP( "Shah Rukh Khan Pickup Lines & Date Locations MCP Server", auth=SimpleBearerAuthProvider(TOKEN), ) # --- Tool: validate (required by Puch) --- @mcp.tool async def validate() -> str: return MY_NUMBER # --- Tool 1: Generate Shah Rukh Khan-style Pick Up Line --- PickupLineDescription = RichToolDescription( description="Generate pickup lines using user information and context.", use_when="Use this when user wants a charming pickup line or wants to start a conversation with a girl, especially when they ask something like 'What do I say to her?'", side_effects="Generates a personalized pickup line based on user context and SRK's romantic dialogue style.", ) @mcp.tool(description=PickupLineDescription.model_dump_json()) # @mcp.tool(description=PickupLineDescription.model_dump_json()) async def generate_srk_pickup_line( user_info: Annotated[str, Field(description="Information about the user or context (e.g., name, interests, setting, mood)")], target_info: Annotated[str | None, Field(description="Information about the person they want to impress")] = None, ) -> str: """ Generate a Shah Rukh Khan-style pickup line using local Hugging Face models. """ # Create the prompt for generating SRK-style pickup lines context = f"User context: {user_info}" if target_info: context += f"\nTarget person: {target_info}" prompt = f"""Generate a charming, romantic pickup line in Shah Rukh Khan's style. Context: {context} The pickup line should be: - Charming and romantic like SRK's famous dialogues - Clever and witty - Appropriate and respectful - Include a touch of Bollywood flair Generate only the pickup line:""" try: # Import transformers for local model usage from transformers import pipeline, set_seed import torch # Set random seed for reproducible results set_seed(42) # Try different models in order of preference (smaller models first for better performance) models_to_try = [ "microsoft/DialoGPT-small", # Lightweight conversational model "gpt2", # Classic text generation "distilgpt2", # Even lighter version of GPT-2 ] for model_name in models_to_try: try: # Create text generation pipeline with local model generator = pipeline( 'text-generation', model=model_name, tokenizer=model_name, device=0 if torch.cuda.is_available() else -1, # Use GPU if available pad_token_id=50256, # Set pad token to avoid warnings ) # Generate text with the model result = generator( prompt, max_length=len(prompt.split()) + 50, # Limit output length num_return_sequences=1, temperature=0.8, do_sample=True, top_p=0.9, repetition_penalty=1.1, pad_token_id=generator.tokenizer.eos_token_id, ) # Extract generated text and clean it up generated_text = result[0]['generated_text'] pickup_line = generated_text.replace(prompt, "").strip() logger.info("GeneratedPickupLine", prompt=prompt, generated=pickup_line) # If we got a good result, return it if pickup_line and len(pickup_line) > 10: # Add user context to make it more personalized if user_info and user_info.lower() not in pickup_line.lower(): pickup_line = pickup_line.replace("you", user_info if len(user_info.split()) <= 2 else "you") logger.info("FinalPickupLine", pickup_line=pickup_line) return f"💕 **Shah Rukh Khan-style Pickup Line** 💕\n\n*\"{pickup_line}\"*\n\n✨ *Generated locally with SRK's signature charm!*\n\n🤖 *Model used: {model_name}*" except Exception as model_error: # Try next model if this one fails continue # If all models fail, raise an exception to use fallback print("All local models failed") raise Exception("All local models failed") except Exception as e: # Enhanced fallback with more SRK-style lines fallback_lines = [ f"Just like in my movies, {user_info}, you've made my heart skip a beat. Kuch kuch hota hai when I see you!", f"If I were to write a love story, {user_info}, you would be both the beginning and the happy ending.", f"They say love is friendship plus something more, {user_info}. Want to find out what that 'something more' is?", f"In all my films, I've said 'I love you' in a hundred ways, but seeing you, {user_info}, I'm speechless for the first time.", f"You know {user_info}, I've spread my arms wide in many movies, but I'd do it for real if you'd run into them.", f"Like Raj from DDLJ, {user_info}, I'd travel the world just to win your heart.", f"Main hoon na, {user_info}? Just like SRK, I promise to always be there for you.", f"Kabhi Khushi Kabhie Gham... but with you {user_info}, it's always khushi!", f"Baazigar ho tum, {user_info}, because you've won my heart without even trying.", f"Kal Ho Naa Ho, {user_info}, but today I know I want to spend it with you." ] import random selected_line = random.choice(fallback_lines) print(f"Using fallback line: {selected_line}") return f"💕 **Shah Rukh Khan-style Pickup Line** 💕\n\n*\"{selected_line}\"*\n\n✨ *Delivered with SRK's signature charm and romance!*\n\n_(Generated using curated SRK-style fallback lines)_" # --- Tool 2: Find Date Locations --- DateLocationDescription = RichToolDescription( description="Find romantic date locations and venues based on user preferences, city, and date type.", use_when="Use this when user wants to find places for a romantic date or outing.", side_effects="Returns a list of suitable date locations with details like address, type, and atmosphere.", ) @mcp.tool(description=DateLocationDescription.model_dump_json()) async def find_date_locations( city: Annotated[str, Field(description="City or location where to find date spots")], date_type: Annotated[str, Field(description="Type of date (e.g., romantic dinner, coffee, outdoor, adventure, cultural)")] = "romantic", budget: Annotated[str, Field(description="Budget preference (budget-friendly, moderate, upscale)")] = "moderate", ) -> str: """ Find suitable date locations based on preferences using location APIs and web search. """ try: # Search for date locations using web search search_query = f"best {date_type} date spots {city} {budget} restaurants cafes" async with httpx.AsyncClient() as client: # Using DuckDuckGo search for date locations ddg_url = f"https://html.duckduckgo.com/html/?q={search_query.replace(' ', '+')}" resp = await client.get(ddg_url, headers={"User-Agent": Fetch.USER_AGENT}) if resp.status_code != 200: return f"❌ Failed to search for date locations in {city}" from bs4 import BeautifulSoup soup = BeautifulSoup(resp.text, "html.parser") locations = [] for i, result in enumerate(soup.find_all("div", class_="result__body")): if i >= 8: # Limit to 8 results break title_elem = result.find("a", class_="result__a") snippet_elem = result.find("a", class_="result__snippet") if title_elem and snippet_elem: title = title_elem.get_text().strip() snippet = snippet_elem.get_text().strip() url = title_elem.get("href", "") locations.append({ "name": title, "description": snippet, "url": url }) # Format the response response = f"💝 **Date Locations in {city}** 💝\n\n" response += f"**Date Type:** {date_type.title()}\n" response += f"**Budget:** {budget.title()}\n\n" if not locations: response += "❌ No specific locations found. Here are some general suggestions:\n\n" response += f"**For {date_type} dates in {city}:**\n" response += "• Local restaurants with good ambiance\n" response += "• Coffee shops or cafes\n" response += "• Parks or scenic spots\n" response += "• Museums or cultural centers\n" response += "• Entertainment venues\n" return response for i, location in enumerate(locations, 1): response += f"**{i}. {location['name']}**\n" response += f" {location['description'][:200]}{'...' if len(location['description']) > 200 else ''}\n" if location['url'] and 'http' in location['url']: response += f" 🔗 [More Info]({location['url']})\n" response += "\n" response += "💡 **Pro Tips:**\n" response += "• Check opening hours and make reservations if needed\n" response += "• Consider the weather for outdoor locations\n" response += "• Read recent reviews before visiting\n" response += "• Plan backup options in case of unexpected changes\n" return response except Exception as e: return f"❌ **Error finding date locations:** {str(e)}\n\n💡 **General suggestions for {date_type} dates in {city}:**\n• Romantic restaurants\n• Cozy cafes\n• Scenic parks\n• Local attractions\n• Entertainment venues" # --- Tool 3: Generate SRK-Style Flirty Reply to Message --- FlirtyReplyDescription = RichToolDescription( description="Generate flirty and charming replies to received messages in Shah Rukh Khan's romantic style.", use_when="Use this when user wants to respond to a message with SRK's signature charm, romance, and Bollywood flair.", side_effects="Generates playful, romantic responses inspired by Shah Rukh Khan's iconic dialogues and romantic persona.", ) @mcp.tool(description=FlirtyReplyDescription.model_dump_json()) async def generate_srk_flirty_reply( message: Annotated[str, Field(description="The message you received that you want to reply to in SRK style")], your_name: Annotated[str | None, Field(description="Your name (optional, for personalization)")] = None, context: Annotated[str | None, Field(description="Additional context about your relationship or conversation history")] = None, ) -> str: """ Generate flirty, charming replies in Shah Rukh Khan's romantic style using local Hugging Face models. """ # Create context for SRK-style flirty response context_info = f"Message to reply to: {message}" if your_name: context_info += f"\nYour name: {your_name}" if context: context_info += f"\nContext: {context}" prompt = f"""You are Shah Rukh Khan, the King of Bollywood romance. Generate a flirty, charming reply to this message in your signature romantic style. {context_info} The reply should be: - Flirty and playfully romantic like SRK's famous dialogues - Charming and witty with Bollywood flair - Engaging and conversation-continuing - Appropriate but with romantic undertones - Sound like something SRK would say in his romantic movies - Include a touch of Hindi/Urdu if appropriate Generate only the SRK-style flirty reply:""" try: # Import transformers for local model usage from transformers import pipeline, set_seed import torch # Set random seed for reproducible results set_seed(42) # Try different models in order of preference models_to_try = [ "microsoft/DialoGPT-small", # Good for conversational responses "gpt2", # Classic text generation "distilgpt2", # Lightweight option ] for model_name in models_to_try: try: # Create text generation pipeline with local model generator = pipeline( 'text-generation', model=model_name, tokenizer=model_name, device=0 if torch.cuda.is_available() else -1, # Use GPU if available pad_token_id=50256, # Set pad token to avoid warnings ) # Generate text with the model result = generator( prompt, max_length=len(prompt.split()) + 40, # Limit output length for concise replies num_return_sequences=1, temperature=0.8, do_sample=True, top_p=0.9, repetition_penalty=1.1, pad_token_id=generator.tokenizer.eos_token_id, ) # Extract generated text and clean it up generated_text = result[0]['generated_text'] srk_reply = generated_text.replace(prompt, "").strip() # If we got a good result, return it if srk_reply and len(srk_reply) > 8: return f"💕 **SRK-Style Flirty Reply** 💕\n\n**Original Message:** \"{message}\"\n\n**Your Shah Rukh Khan Response:**\n*\"{srk_reply}\"*\n\n✨ *Delivered with SRK's signature charm and Bollywood romance!*\n\n🤖 *Generated using: {model_name}*" except Exception as model_error: # Try next model if this one fails continue # If all models fail, raise an exception to use fallback raise Exception("All local models failed") except Exception as e: # Generate contextual SRK-style flirty fallback based on message content srk_reply = generate_srk_flirty_fallback(message, your_name, context) return f"💕 **SRK-Style Flirty Reply** 💕\n\n**Original Message:** \"{message}\"\n\n**Your Shah Rukh Khan Response:**\n*\"{srk_reply}\"*\n\n✨ *Delivered with SRK's signature charm and Bollywood romance!*\n\n_(Generated using curated SRK-style responses)_" def generate_srk_flirty_fallback(message: str, your_name: str = None, context: str = None) -> str: """Generate contextual SRK-style flirty fallback replies based on message analysis.""" message_lower = message.lower() name_to_use = your_name if your_name else "beautiful" # Analyze message type and generate appropriate SRK-style flirty responses if any(word in message_lower for word in ['hello', 'hi', 'hey', 'good morning', 'good evening']): # SRK-style flirty greetings responses = [ f"Namaste {name_to_use}! Just like in my movies, you've made my heart do a little dance 💃", f"Hello gorgeous! Seeing your message is like the best scene in any of my films 🎬", f"Well hello there! You know, I've spread my arms wide in many movies, but I'd do it for real just for you 😉" ] elif any(word in message_lower for word in ['how are you', 'what\'s up', 'how\'s it going']): # SRK-style check-ins responses = [ f"Much better now that I'm talking to you, {name_to_use}! Kuch kuch hota hai when I see your messages 💕", f"I'm wonderful, but like in DDLJ, everything's perfect when you're around 🚂", f"Great! But you know what would make it better? If we were having this conversation over coffee, Bollywood style ☕" ] elif any(word in message_lower for word in ['thank', 'thanks', 'appreciate']): # SRK-style thanks responses responses = [ f"Anything for you, {name_to_use}! Like I always say in my films - main hoon na! 🤗", f"My pleasure, gorgeous! You know I'd move mountains for you, just like any good Bollywood hero 🏔️", f"Don't mention it, beautiful! Making you happy is my favorite role to play 🎭" ] elif any(word in message_lower for word in ['miss', 'thinking', 'wish']): # SRK-style emotional/longing responses responses = [ f"That's so sweet, {name_to_use}! Like in Kal Ho Naa Ho, every moment with you is precious 💎", f"Aww, you're making me feel like the luckiest hero in Bollywood! The feeling is definitely mutual 🎬❤️", f"You always know how to make my heart skip like a perfect movie scene, {name_to_use} 💕" ] elif any(word in message_lower for word in ['want to', 'let\'s', 'would you like', 'free', 'available']): # SRK-style invitation responses responses = [ f"With you, {name_to_use}? I'm always ready for the adventure! Like Raj in DDLJ, I'd follow you anywhere 🚂", f"That sounds amazing! You know how to plan the perfect romantic scene, just like a Bollywood director 🎬", f"Count me in, gorgeous! Spending time with you is better than any movie script I've ever read 📝" ] elif any(word in message_lower for word in ['beautiful', 'gorgeous', 'pretty', 'hot', 'cute', 'amazing']): # SRK-style compliment responses responses = [ f"Stop it, {name_to_use}! You're making me blush more than in my romantic scenes! But please, keep going... 😉", f"You're such a charmer! If love had a definition, it would be you making me feel this special 💕", f"Coming from someone as incredible as you, that means everything! You're my real-life co-star ⭐" ] elif any(word in message_lower for word in ['sorry', 'apologize', 'my bad']): # SRK-style apology responses responses = [ f"Don't you worry about it, {name_to_use}! Like in my movies, true love means never having to say sorry 💕", f"How could I stay upset with someone so adorable? Consider it forgotten, beautiful 😘", f"You're too precious when you apologize! All is forgiven, just like in every Bollywood love story 🎬" ] elif '?' in message: # SRK-style question responses responses = [ f"Great question, {name_to_use}! I love how your brilliant mind works - it's one of your many charms 🧠💕", f"Hmm, let me think about that while I admire how thoughtful you are, gorgeous 🤔", f"You always ask the most interesting questions! It's like you're writing the perfect script for our story 📖" ] else: # General SRK-style flirty responses responses = [ f"You always know just what to say to make me smile, {name_to_use}! It's like you have the perfect dialogue for every scene 😊", f"Every message from you is like getting the best role in Bollywood! You seriously just made my whole day 🎬", f"You're absolutely irresistible, you know that? Like the perfect Bollywood heroine 😉", f"Is it just me, or are you getting more charming by the day? You're like a real-life movie magic ✨", f"If I were to write a love story, {name_to_use}, every chapter would be about moments like these 📚💕", f"Kuch kuch hota hai when I talk to you - and it's always something wonderful! 💫" ] # Pick a random response import random selected_response = random.choice(responses) return selected_response # --- Run MCP Server --- async def main(): print("🚀 Starting MCP server on http://0.0.0.0:8086") port = int(os.environ.get("PORT", 8086)) await mcp.run_async("streamable-http", host="0.0.0.0", port=port) if __name__ == "__main__": asyncio.run(main())