app.py

import gradio as gr
import requests
from typing import List, Dict, Optional
from huggingface_hub import HfApi
import os
from dotenv import load_dotenv
import csv
from pinecone import Pinecone
from openai import OpenAI

# Load environment variables
load_dotenv()

# Initialize HF API with token if available
HF_TOKEN = os.getenv("HF_TOKEN")
api = HfApi(token=HF_TOKEN) if HF_TOKEN else HfApi()

def keyword_search_hf_spaces(query: str = "", limit: int = 3) -> Dict:
    """
    Search for MCPs in Hugging Face Spaces.
    
    Args:
        query: Search query string
        limit: Maximum number of results to return (default: 3)
        
    Returns:
        Dictionary containing search results with MCP information
    """
    try:
        print(f"Debug - Search query: '{query}'")  # Debug log
        
        # Use list_spaces API with mcp-server filter and sort by likes
        spaces = list(api.list_spaces(
            search=query,
            sort="likes",
            direction=-1,  # Descending order
            filter="mcp-server"
        ))
        
        results = []
        for space in spaces[:limit]:  # Process up to limit matches
            try:
                space_info = {
                    "id": space.id,
                    "likes": space.likes,
                    "trending_score": space.trending_score,
                    "source": "huggingface"
                }
                results.append(space_info)
            except Exception as e:
                print(f"Error processing space {space.id}: {str(e)}")
                continue
        
        return {
            "results": results,
            "total": len(results)
        }
    except Exception as e:
        print(f"Debug - Critical error in keyword_search_hf_spaces: {str(e)}")
        return {
            "error": str(e),
            "results": [],
            "total": 0
        }

def keyword_search_smithery(query: str = "", limit: int = 3) -> Dict:
    """
    Search for MCPs in Smithery Registry.
    
    Args:
        query: Search query string
        limit: Maximum number of results to return (default: 3)
        
    Returns:
        Dictionary containing search results with MCP information
    """
    try:
        # Get Smithery token from environment
        SMITHERY_TOKEN = os.getenv("SMITHERY_TOKEN")
        if not SMITHERY_TOKEN:
            return {
                "error": "SMITHERY_TOKEN not found",
                "results": [],
                "total": 0
            }
        
        # Prepare headers and query parameters
        headers = {
            'Authorization': f'Bearer {SMITHERY_TOKEN}'
        }
        
        # Add filters for deployed and verified servers
        search_query = f"{query} is:deployed"
        
        params = {
            'q': search_query,
            'page': 1,
            'pageSize': 100  # Get maximum results
        }
        
        # Make API request
        response = requests.get(
            'https://registry.smithery.ai/servers',
            headers=headers,
            params=params
        )
        
        if response.status_code != 200:
            return {
                "error": f"Smithery API error: {response.status_code}",
                "results": [],
                "total": 0
            }
        
        # Parse response
        data = response.json()
        results = []
        
        # Sort servers by useCount and take top results up to limit
        servers = sorted(data.get('servers', []), key=lambda x: x.get('useCount', 0), reverse=True)[:limit]
        
        for server in servers:
            server_info = {
                "id": server.get('qualifiedName'),
                "name": server.get('displayName'),
                "description": server.get('description'),
                "likes": server.get('useCount', 0),
                "source": "smithery"
            }
            results.append(server_info)
        
        return {
            "results": results,
            "total": len(results)
        }
        
    except Exception as e:
        return {
            "error": str(e),
            "results": [],
            "total": 0
        }

def keyword_search(query: str, sources: List[str], limit: int = 3) -> Dict:
    """
    Search for MCPs using keyword matching.
    
    Args:
        query: Keyword search query
        sources: List of sources to search from ('huggingface', 'smithery')
        limit: Maximum number of results to return (default: 3)
        
    Returns:
        Dictionary containing combined search results
    """
    all_results = []
    
    if "huggingface" in sources:
        hf_results = keyword_search_hf_spaces(query, limit)
        all_results.extend(hf_results.get("results", []))
    
    if "smithery" in sources:
        smithery_results = keyword_search_smithery(query, limit)
        all_results.extend(smithery_results.get("results", []))
    
    return {
        "results": all_results,
        "total": len(all_results),
        "search_type": "keyword"
    }

def embedding_search_hf_spaces(query: str = "", limit: int = 3) -> Dict:
    """
    Search for MCPs in Hugging Face Spaces using semantic embedding matching.
    
    Args:
        query: Natural language search query
        limit: Maximum number of results to return (default: 3)
        
    Returns:
        Dictionary containing search results with MCP information
    """
    try:
        # Initialize Pinecone and OpenAI
        pinecone_api_key = os.getenv('PINECONE_API_KEY')
        openai_api_key = os.getenv('OPENAI_API_KEY')
        
        if not pinecone_api_key or not openai_api_key:
            return {
                "error": "API keys not found",
                "results": [],
                "total": 0
            }
        
        # Initialize clients
        pc = Pinecone(api_key=pinecone_api_key)
        index = pc.Index("hf-mcp")
        client = OpenAI(api_key=openai_api_key)
        
        # Generate embedding using OpenAI
        response = client.embeddings.create(
            input=query,
            model="text-embedding-3-large"
        )
        query_embedding = response.data[0].embedding
        
        # Search in Pinecone using the generated embedding
        results = index.query(
            namespace="",
            vector=query_embedding,
            top_k=limit
        )
        
        # Process results and get detailed information
        space_results = []
        if not results.matches:
            return {
                "results": [],
                "total": 0
            }
        
        for match in results.matches:
            space_id = match.id
            try:
                # Remove 'spaces/' prefix if present
                repo_id = space_id.replace('spaces/', '')
                
                # Get space information from HF API
                space = api.space_info(repo_id)
                space_info = {
                    "id": space.id,
                    "likes": space.likes,
                    "trending_score": space.trending_score,
                    "source": "huggingface",
                    "score": match.score  # Add similarity score
                }
                space_results.append(space_info)
            except Exception as e:
                continue
        
        return {
            "results": space_results,
            "total": len(space_results)
        }
        
    except Exception as e:
        return {
            "error": str(e),
            "results": [],
            "total": 0
        }

def embedding_search_smithery(query: str = "", limit: int = 3) -> Dict:
    """
    Search for MCPs in Smithery Registry using semantic embedding matching.
    
    Args:
        query: Natural language search query
        limit: Maximum number of results to return (default: 3)
        
    Returns:
        Dictionary containing search results with MCP information
    """
    try:
        # Initialize Pinecone and OpenAI
        from pinecone import Pinecone
        from openai import OpenAI
        import os
        
        pinecone_api_key = os.getenv('PINECONE_API_KEY')
        openai_api_key = os.getenv('OPENAI_API_KEY')
        smithery_token = os.getenv('SMITHERY_TOKEN')
        
        if not pinecone_api_key or not openai_api_key or not smithery_token:
            return {
                "error": "API keys not found",
                "results": [],
                "total": 0
            }
        
        # Initialize clients
        pc = Pinecone(api_key=pinecone_api_key)
        index = pc.Index("smithery-mcp")
        client = OpenAI(api_key=openai_api_key)
        
        # Generate embedding using OpenAI
        response = client.embeddings.create(
            input=query,
            model="text-embedding-3-large"
        )
        query_embedding = response.data[0].embedding
        
        # Search in Pinecone using the generated embedding
        results = index.query(
            namespace="",
            vector=query_embedding,
            top_k=limit
        )
        
        # Process results and get detailed information from Smithery
        server_results = []
        if not results.matches:
            return {
                "results": [],
                "total": 0
            }
        
        # Prepare headers for Smithery API
        headers = {
            'Authorization': f'Bearer {smithery_token}'
        }
        
        for match in results.matches:
            server_id = match.id
            try:
                # Get server information from Smithery API
                response = requests.get(
                    f'https://registry.smithery.ai/servers/{server_id}',
                    headers=headers
                )
                
                if response.status_code != 200:
                    continue
                
                server = response.json()
                server_info = {
                    "id": server.get('qualifiedName'),
                    "name": server.get('displayName'),
                    "description": server.get('description'),
                    "likes": server.get('useCount', 0),
                    "source": "smithery",
                    "score": match.score  # Add similarity score
                }
                server_results.append(server_info)
            except Exception as e:
                continue
        
        return {
            "results": server_results,
            "total": len(server_results)
        }
        
    except Exception as e:
        return {
            "error": str(e),
            "results": [],
            "total": 0
        }

def embedding_search(query: str, sources: List[str], limit: int = 3) -> Dict:
    """
    Search for MCPs using semantic embedding matching.
    
    Args:
        query: Natural language search query
        sources: List of sources to search from ('huggingface', 'smithery')
        limit: Maximum number of results to return (default: 3)
        
    Returns:
        Dictionary containing combined search results
    """
    all_results = []
    
    if "huggingface" in sources:
        try:
            hf_results = embedding_search_hf_spaces(query, limit)
            all_results.extend(hf_results.get("results", []))
        except Exception as e:
            # Fallback to keyword search if vector search fails
            hf_results = keyword_search_hf_spaces(query, limit)
            all_results.extend(hf_results.get("results", []))
    
    if "smithery" in sources:
        try:
            smithery_results = embedding_search_smithery(query, limit)
            all_results.extend(smithery_results.get("results", []))
        except Exception as e:
            # Fallback to keyword search if vector search fails
            smithery_results = keyword_search_smithery(query, limit)
            all_results.extend(smithery_results.get("results", []))
    
    return {
        "results": all_results,
        "total": len(all_results),
        "search_type": "embedding"
    }

# Create the Gradio interface
with gr.Blocks(title="🚦 Router MCP", css="""
    #client_radio {
        margin-top: 0 !important;
        padding-top: 0 !important;
    }
    #client_radio .radio-group {
        gap: 0.5rem !important;
    }
""") as demo:
    gr.Markdown("# 🚦 Router MCP")
    gr.Markdown("### Search MCP compatible spaces using natural language")
    
    with gr.Row():
        with gr.Column():
            query_input = gr.Textbox(
                label="Describe the MCP Server you're looking for",
                placeholder="e.g., 'I need an MCP Server that can generate images'"
            )
            
            gr.Markdown("### Select sources to search")
            hf_checkbox = gr.Checkbox(label="Hugging Face Spaces", value=True)
            smithery_checkbox = gr.Checkbox(label="Smithery", value=False)
            registry_checkbox = gr.Checkbox(label="Registry (Coming Soon)", value=False, interactive=False)
            
            result_limit = gr.Number(
                label="Maximum number of results for each source",
                value=3,
                minimum=1,
                maximum=20,
                step=1
            )
            
            gr.Markdown("### Select your MCP Client")
            client_radio = gr.Radio(
                choices=["Cursor", "Windsurf", "Claude Desktop", "VS Code", "Gradio"],
                label="",
                value="Cursor",
                interactive=True,
                elem_id="client_radio"
            )
            
            with gr.Row():
                keyword_search_button = gr.Button("Keyword Search")
                embedding_search_button = gr.Button("Semantic Search")
        
        with gr.Column():
            results_output = gr.JSON(label="Search Results")
    
    # Set up event handlers
    def get_sources():
        return ["huggingface" if hf_checkbox.value else "", "smithery" if smithery_checkbox.value else ""]
    
    def handle_keyword_mcp_search(query: str, hf: bool, sm: bool, limit: int) -> Dict:
        """
        Handle keyword-based search for MCP servers across selected sources. If the client (such as Cursor or Claude) encounters a task it cannot handle, it can use MCP to search for other tools that can.

        Args:
            query (str): The search query string to find matching MCP servers
            hf (bool): Whether to include Hugging Face Spaces in the search (converted to "huggingface" string if True)
            sm (bool): Whether to include Smithery in the search (converted to "smithery" string if True)
            limit (int): Maximum number of results to return per source (default: 3)

        Returns:
            Dict: A dictionary containing the search results with the following keys:
                - results: List of found MCP servers
                - total: Total number of results
                - search_type: Type of search performed ("keyword")
        """
        return keyword_search(
            query,
            ["huggingface" if hf else "", "smithery" if sm else ""],
            int(limit)
        )
    
    def handle_embedding_mcp_search(query: str, hf: bool, sm: bool, limit: int) -> Dict:
        """
        Handle semantic embedding-based search for MCP servers across selected sources. If the client (such as Cursor or Claude) encounters a task it cannot handle, it can use MCP to search for other tools that can.

        Args:
            query (str): The natural language search query to find semantically similar MCP servers
            hf (bool): Whether to include Hugging Face Spaces in the search (converted to "huggingface" string if True)
            sm (bool): Whether to include Smithery in the search (converted to "smithery" string if True)
            limit (int): Maximum number of results to return per source (default: 3)

        Returns:
            Dict: A dictionary containing the search results with the following keys:
                - results: List of found MCP servers with similarity scores
                - total: Total number of results
                - search_type: Type of search performed ("embedding")
        """
        return embedding_search(
            query,
            ["huggingface" if hf else "", "smithery" if sm else ""],
            int(limit)
        )
    
    keyword_search_button.click(
        fn=handle_keyword_mcp_search,
        inputs=[query_input, hf_checkbox, smithery_checkbox, result_limit],
        outputs=results_output
    )
    
    embedding_search_button.click(
        fn=handle_embedding_mcp_search,
        inputs=[query_input, hf_checkbox, smithery_checkbox, result_limit],
        outputs=results_output
    )
    
    # query_input.submit(
    #     fn=handle_embedding_search,
    #     inputs=[query_input, hf_checkbox, smithery_checkbox, result_limit],
    #     outputs=results_output
    # )

if __name__ == "__main__":
    demo.launch(mcp_server=True)