Spaces:
Running
Running
| import json | |
| from typing import Literal | |
| from datetime import datetime | |
| import gradio as gr | |
| from huggingface_hub import list_models, model_info, hf_hub_download | |
| def search_models( | |
| search: str = None, | |
| library: str = None, | |
| tags: str = None, | |
| pipeline_tag: str = None, | |
| sort: Literal[ | |
| "trending_score", "last_modified", "created_at", "downloads", "likes" | |
| ] = "trending_score", | |
| direction: Literal["descending", "ascending"] = "descending", | |
| limit: int = 20, | |
| ) -> str: | |
| """ | |
| Search models on Hugging Face Hub. | |
| Use this tool to search for models by name, tags, or other filters, and to get a list of model IDs. | |
| This is the first step when you need to find a specific model before retrieving its details. | |
| Parameters: | |
| search (str, optional): A string to search for in model IDs or names (e.g., "deepseek"). | |
| library (list[str], optional): List of libraries the models use (e.g., ["pytorch", "tensorflow"]). | |
| tags (list[str], optional): List of tags to filter models by (e.g., ["text-generation", "llama"]). | |
| pipeline_tag (str, optional): Filter by pipeline tag (e.g., "text-generation"). | |
| sort (Literal["trending_score", "last_modified", "created_at", "downloads", "likes"], default="trending_score"): Sort models by the specified key. | |
| direction (int, default=-1): Sort direction: -1 for descending, 1 for ascending. | |
| limit (int, default=20): Maximum number of models to return. | |
| Returns: | |
| list[str]: A list of model IDs matching the search criteria. | |
| Examples: | |
| - To find trending models: search_models(sort="trending_score", limit=10) | |
| - To search for models related to "deepseek": search_models(search="deepseek", sort="likes", limit=5) | |
| - To filter by tag: search_models(tags=["text-generation"], pipeline_tag="text-generation") | |
| """ | |
| try: | |
| library = library.split(",") if library else None | |
| tags = tags.split(",") if tags else None | |
| direction = -1 if direction == "descending" else 1 | |
| models = list_models( | |
| library=library, | |
| tags=tags, | |
| search=search, | |
| pipeline_tag=pipeline_tag, | |
| sort=sort, | |
| direction=direction, | |
| limit=limit, | |
| ) | |
| return json.dumps([model.modelId for model in models]) | |
| except Exception as e: | |
| return f"Error: {e}" | |
| def get_model_info(model_id: str) -> dict: | |
| """ | |
| Get structured metadata about a model on the Hugging Face Hub. | |
| Use this when you need specific fields like downloads, tags, or other metadata. | |
| For comprehensive model information, use `get_model_card`. | |
| This tool requires the exact model ID, which can be obtained using `search_models`. | |
| If you have a partial name or tag, use `search_models` first to find the exact ID. | |
| Parameters: | |
| model_id (str): The exact model ID in the format "organization/model-name" (e.g., "DeepSeek/DeepSeek-R1"). | |
| Returns: | |
| dict: A dictionary containing model information including available fields such as: | |
| - id: The model ID | |
| - author: The author of the model | |
| - created_at: The creation date | |
| - last_modified: The last modified date | |
| - downloads: Number of downloads | |
| - likes: Number of likes | |
| - tags: List of tags | |
| - pipeline_tag: The pipeline tag | |
| - library_name: The library name | |
| - license: The model license | |
| - base_model: The base model (if available) | |
| - siblings: List of repository files (if available) | |
| - datasets: Datasets used to train the model (if available) | |
| - spaces: List of spaces using this model (if available) | |
| - xet_enabled: Whether XET is enabled (if available) | |
| Raises: | |
| Exception: If the model_id is invalid or not found. Use search_models to find the correct ID. | |
| Example: | |
| - First, find the model ID: search_models(search="deepseek", sort="likes", limit=1) | |
| - Then, get the model info: get_model_info("DeepSeek/DeepSeek-R1") | |
| """ | |
| try: | |
| model = model_info(model_id) | |
| result = {} | |
| if hasattr(model, "id") and model.id is not None: | |
| result["id"] = model.id | |
| if hasattr(model, "author") and model.author is not None: | |
| result["author"] = model.author | |
| if hasattr(model, "created_at") and model.created_at is not None: | |
| result["created_at"] = str(model.created_at) | |
| if hasattr(model, "last_modified") and model.last_modified is not None: | |
| result["last_modified"] = str(model.last_modified) | |
| if hasattr(model, "downloads") and model.downloads is not None: | |
| result["downloads"] = model.downloads | |
| if hasattr(model, "likes") and model.likes is not None: | |
| result["likes"] = model.likes | |
| if hasattr(model, "tags") and model.tags is not None: | |
| result["tags"] = model.tags | |
| if hasattr(model, "pipeline_tag") and model.pipeline_tag is not None: | |
| result["pipeline_tag"] = model.pipeline_tag | |
| if hasattr(model, "library_name") and model.library_name is not None: | |
| result["library_name"] = model.library_name | |
| if hasattr(model, "card_data") and model.card_data is not None: | |
| if ( | |
| hasattr(model.card_data, "license") | |
| and model.card_data.license is not None | |
| ): | |
| result["license"] = model.card_data.license | |
| if ( | |
| hasattr(model.card_data, "base_model") | |
| and model.card_data.base_model is not None | |
| ): | |
| result["base_model"] = model.card_data.base_model | |
| if ( | |
| hasattr(model.card_data, "datasets") | |
| and model.card_data.datasets is not None | |
| ): | |
| result["datasets"] = model.card_data.datasets | |
| if hasattr(model, "siblings") and model.siblings is not None: | |
| result["siblings"] = [] | |
| for s in model.siblings: | |
| if isinstance(s, str): | |
| result["siblings"].append(s) | |
| else: | |
| result["siblings"].append({ | |
| k: str(v) if isinstance(v, datetime) else v | |
| for k, v in s.__dict__.items() if not k.startswith('_') | |
| }) | |
| if hasattr(model, "spaces") and model.spaces is not None: | |
| result["spaces"] = [] | |
| for s in model.spaces: | |
| if isinstance(s, str): | |
| result["spaces"].append(s) | |
| else: | |
| result["spaces"].append({ | |
| k: str(v) if isinstance(v, datetime) else v | |
| for k, v in s.__dict__.items() if not k.startswith('_') | |
| }) | |
| if hasattr(model, "xet_enabled") and model.xet_enabled is not None: | |
| result["xet_enabled"] = model.xet_enabled | |
| return json.dumps(result) | |
| except Exception as e: | |
| return f"Error: {e}" | |
| def get_model_card(model_id: str) -> str: | |
| """ | |
| Get the complete model card (README.md) for a specific model on Hugging Face Hub. | |
| Use this when you need comprehensive model documentation including usage examples, model limitations, etc. | |
| For only structured metadata, use `get_model_info` instead. | |
| This tool requires the exact model ID, which can be obtained using `search_models`. | |
| If you have a partial name or tag, use `search_models` first to find the exact ID. | |
| Args: | |
| model_id (str): The model ID in the format "organization/model-name" (e.g., "DeepSeek/DeepSeek-R1"). | |
| Returns: | |
| str: The markdown content of the model card. | |
| Example: | |
| - First, find the model ID: search_models(search="deepseek", sort="likes", limit=1) | |
| - Then, get the model card: get_model_card("DeepSeek/DeepSeek-R1") | |
| """ | |
| try: | |
| filepath = hf_hub_download(model_id, "README.md") | |
| with open(filepath, "r", encoding="utf-8") as f: | |
| content = f.read() | |
| return content | |
| except Exception as e: | |
| return f"Error: {e}" | |
| description_html = """ | |
| <h1>🤗 Hugging Face MCP Server</h1> | |
| <p>Use AI Agents to interact with the Hugging Face Hub.</p> | |
| <h3>Available tools:</h3> | |
| <ul> | |
| <li><strong>search_models</strong>: Find models by name, tags, etc.</li> | |
| <li><strong>get_model_info</strong>: Get model metadata</li> | |
| <li><strong>get_model_card</strong>: View model documentation</li> | |
| </ul> | |
| <h3>Setup (for Cursor):</h3> | |
| <ol> | |
| <li>Click <strong>"Use via API"</strong> → <strong>"MCP"</strong> tab</li> | |
| <li>Copy SSE config: | |
| <pre> | |
| { | |
| "mcpServers": { | |
| "gradio": { | |
| "url": "https://dylanebert-huggingface-mcp.hf.space/gradio_api/mcp/sse" | |
| } | |
| } | |
| } | |
| </pre> | |
| </li> | |
| <li>Paste in: <strong>Cursor Settings</strong> → <strong>MCP</strong> → <strong>Add Global MCP Server</strong></li> | |
| <li>Chat in agent mode with prompts like: "list top 10 trending models on Hugging Face"</li> | |
| </ol> | |
| <hr> | |
| """ | |
| search_models = gr.Interface( | |
| fn=search_models, | |
| description=description_html, | |
| inputs=[ | |
| gr.Textbox(label="search", value=""), | |
| gr.Textbox(label="library", value=""), | |
| gr.Textbox(label="tags", value=""), | |
| gr.Textbox(label="pipeline_tag", value=""), | |
| gr.Radio(label="sort", choices=["trending_score", "last_modified", "created_at", "downloads", "likes"], value="trending_score"), | |
| gr.Radio(label="direction", choices=["descending", "ascending"], value="descending"), | |
| gr.Number(label="limit", value=20), | |
| ], | |
| outputs="text") | |
| get_model_info = gr.Interface( | |
| fn=get_model_info, | |
| description=description_html, | |
| inputs=[ | |
| gr.Textbox(label="model_id", value=""), | |
| ], | |
| outputs="text") | |
| get_model_card = gr.Interface( | |
| fn=get_model_card, | |
| description=description_html, | |
| inputs=[ | |
| gr.Textbox(label="model_id", value=""), | |
| ], | |
| outputs="text") | |
| demo = gr.TabbedInterface( | |
| interface_list=[search_models, get_model_info, get_model_card], | |
| tab_names=["search_models", "get_model_info", "get_model_card"] | |
| ) | |
| demo.launch(mcp_server=True) | |