What does this skill do, and when should you use it?
This skill connects an agent to the Hugging Face Hub through the Hugging Face MCP server. It searches models, datasets, Spaces, and papers, retrieves repository details and documentation, and runs Hugging Face Jobs. It can also use MCP-enabled Gradio Spaces as callable AI tools. The skill is available only when the agent is connected to the HF MCP server.
It calls model_search, dataset_search, space_search, and paper_search to discover Hub resources; uses hub_repo_details to retrieve repository details and READMEs; searches and fetches library documentation with hf_doc_search and hf_doc_fetch; submits, inspects, and reads logs for jobs through hf_jobs; discovers, inspects, and invokes Gradio Spaces with dynamic_space; checks identity with hf_whoami; and can call gr1_flux1_schnell_infer for image generation.
- A developer choosing a model for code generation can search by task, author, downloads, or trending score and inspect model documentation.
- A data team seeking English sentiment-analysis data can search datasets using language and task-category tags and review repository documentation.
- A developer who wants to reuse an existing AI application can find an MCP-enabled Gradio Space, inspect its parameters, and invoke it.
- An engineer validating GPU access or running training code can submit Hugging Face Jobs and inspect status and logs.
- A developer learning PEFT or LoRA workflows can search and fetch Hugging Face library documentation.
- A researcher studying reinforcement learning or related topics can search papers and inspect linked Hub resources.
How do you install this skill?
- Before using hf_jobs or dynamic_space, obtain explicit user authorization and review scripts, images, parameters, input files, and secrets to avoid sending sensitive data to remote services.
- Core functionality depends on the Hugging Face MCP server and overseas network services; reachability from mainland-China networks may be limited, so provide an accessible or local fallback.
- Add guidance for authentication and tool failures, cost controls, cancellation/rollback, output formats, and tests covering the key tool paths.
- Network access
- MCP Server
Hugging Face MCP server
First configure and connect the Hugging Face MCP server at https://huggingface.co/settings/mcp. The repository README says to copy or symlink hf-mcp/skills/hf-mcp/ into a standard Codex .agents/skills location, such as $REPO_ROOT/.agents/skills or $HOME/.agents/skills. The source does not provide a dedicated installation command for this individual skill.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/huggingface/skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/hf-mcp/skills/hf-mcp" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
With the skill loaded and the MCP server connected, give the agent a task such as Find the best model for code generation, Find datasets for sentiment analysis in English, What's happening with my training job?, or Create an image of a robot reading a book. The skill selects the relevant Hugging Face MCP tools.
What are this skill's strengths and limitations?
- Covers model, dataset, Space, paper, documentation, and compute-job operations.
- Can retrieve repository READMEs for fuller model or dataset context.
- Can discover and invoke Gradio Spaces as AI tools.
- Includes examples for submitting jobs, checking status, reading logs, and scheduling recurring jobs.
- Requires an active connection to the Hugging Face MCP server; otherwise it is unavailable.
- The source does not document pricing, permission scopes, or retry behavior.
- The source provides no test-suite or platform-testing evidence for this skill.
- It focuses on MCP tool calls rather than providing the full Hugging Face CLI workflow.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Hugging Face MCP Assistant this page | 36 · Not recommended | ★ 11k | 18d ago | Apache-2.0 |
| Hugging Face CLI Skill | 55 · Use with care | ★ 38k | 14d ago | MIT |
| Hugging Face ZeroGPU Builder | 56 · Use with care | ★ 11k | 18d ago | Apache-2.0 |
| HF Paper Publisher | 45 · Not recommended | ★ 38k | 14d ago | MIT |
| Hugging Face Local Model Runner | 44 · Not recommended | ★ 11k | 18d ago | Apache-2.0 |
The README presents hf-cli as the recommended first skill for core Hub operations, while hf-mcp exposes search, documentation, Spaces, and Jobs capabilities through MCP server tools. The source does not provide comparisons with other competitors.
How did FollowSkills review this skill?
The skill explicitly depends on a connected Hugging Face MCP server and includes cloud jobs, dynamic Space invocation, and HF_TOKEN secrets. It does not specify least privilege, user confirmation, sensitive-data handling, data flows, rollback, or controls for external side effects, so substantial points are deducted. No credential theft, malware, or destructive default is evident.
Tool selection and examples are broadly internally consistent, and the dependency on the HF MCP server is stated. However, there is no guidance for unavailable tools, authentication failures, invalid parameters, failed remote jobs, abnormal input, or diagnostic failure feedback, and no committed tests cover the skill's key paths. Static calibration limits this to 10 or below.
The document identifies concrete scenarios across models, datasets, Spaces, papers, documentation, and jobs for AI assistants using MCP. Trigger semantics, non-fit boundaries, input/output contracts, environment compatibility, Chinese support, and mainland-China reachability are insufficiently specified; core functionality depends on a remote Hugging Face service.
The SKILL.md has clear frontmatter, use cases, a tool-selection table, and tips. Repository context supplies Apache-2.0 licensing, generated-artifact CI, security reporting, and publishing guidance. The skill itself lacks installation and dependency details, versioning/changelog information, known limitations, explicit maintenance ownership, and a concrete update path.
Examples map directly to searches, repository details, documentation retrieval, Space invocation, and job management, providing plausible tool-assisted value. Nevertheless, output formats, success criteria, failure handling, and representative usable results are unspecified, while authentication, remote-service dependence, and cost implications are not explained. Static evidence supports only limited core-task completion, capped at 7.
The source provides auditable tool-call examples, and repository CI validates generated artifacts and the skills index. There are no skill-specific tests, execution records, or independent corroboration for the key paths, so conclusions rely mainly on documentation claims. Static calibration limits this to 5 or below.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →