Search and Deploy AI Models from Hugging Face Hub

Hugging Face MCP Tools guides an assistant connected to the HF MCP server through Hub discovery, documentation retrieval, Space invocation, and compute-job workflows. The skill's tool-selection guide maps the user's goal to the appropriate search, repository, documentation, or execution tool.

Model search can filter by task, author, or query and sort by trending score or downloads. Examples find code-generation candidates, compare model providers, inspect a model card, explore current trends, and locate quantized GGUF variants. Repository details can include the README so selection is informed by model documentation rather than search names alone.

Dataset search combines a topic with language and task-category tags, then retrieves repository details using the dataset type. Paper search supports topic research and can lead into details of a linked model. Documentation workflows search a product's library guides and fetch the selected page, with LoRA fine-tuning through PEFT as the example.

Space discovery filters for MCP-enabled applications. Before invoking an app, the workflow views its accepted parameters, then sends the corresponding inputs. Space applications support background removal, audio transcription, and image generation through available Space tasks or the named image tool.

Use this skill when the assistant has an HF MCP connection and needs models, datasets, papers, library documentation, Gradio applications, or compute services. It fits comparative research as well as moving from a discovered resource into its documented use.

Keep the workflow within the upstream product and API scope. Before executing jobs or applying changes, verify current behavior, pricing, quotas, credentials, and deployment effects against official documentation. Example calls do not replace environment-specific checks, security review, or approval for costly and destructive actions.

Inputs include search queries, task filters, author names, dataset tags, repository identifiers, documentation products, Space names and parameters, and compute scripts or container commands. Outputs include search results, model or dataset READMEs, fetched documentation, Space results, and compute-job records.

Search model listings with a natural-language query:

model_search(sort="trendingScore", limit=20)

Compute examples run a Python script using a UV job or launch a training command in a container with a selected GPU flavor. Private-repository jobs receive the HF token through job secrets. Monitoring lists jobs and retrieves logs by job ID; recurring execution supplies a cron schedule for a scheduled UV operation.

The tool set connects Hugging Face Hub, its documentation, Gradio Spaces, and CPU or GPU jobs. Dynamic Space discovery lists available tasks, and hf_whoami checks authentication. Use Hugging Face's MCP settings for server setup.