Hugging Face is the open hub for machine-learning models, datasets, Spaces, and research papers, and it exposes a public MCP server over that catalogue. Connecting it to clariBI lets you search and compare the open-ML landscape inside an analysis, with no account or key required for anonymous use.
Why connect Hugging Face
Comparing models, tracking which datasets a paper uses, or pulling repository metadata by hand means jumping between many Hub pages and copying numbers out.
With Hugging Face connected, you can ask "Which text-generation models from a given org are trending and how large are they?", "Find recent papers on retrieval-augmented generation", or "Compare downloads and likes across these five models" and the AI engine reads the public Hub live.
How the connection works
clariBI talks to Hugging Face through its hosted MCP server at https://huggingface.co/mcp. Authentication uses no authentication, because this is a public read-only endpoint. No credentials are sent or stored.
sequenceDiagram
actor U as You
participant C as clariBI
participant V as Hugging Face
U->>C: Click Authorize with Hugging Face
C->>V: Open OAuth authorization
V-->>U: Grant read access?
U->>V: Approve
V-->>C: Authorization code
C->>V: Exchange code for tokens
V-->>C: Access + refresh tokens
C->>C: Encrypt and store credentials
C-->>U: Connection ready
Available tools
clariBI exposes the read-only Hugging Face tools that the vendor's MCP server publishes at connection time. Write operations (create, update, delete, send, refund) are filtered out by a name-pattern blocklist before any tool reaches the analysis engine, so connecting Hugging Face cannot modify data on the vendor side.
The exact tool inventory depends on the Hugging Face features your account has access to. After connecting, try a few natural-language questions to see what Hugging Face data clariBI can pull.
Data flow during analysis
When you ask a question that maps to Hugging Face, the AI engine routes to the right tool, reads the result, and pairs the answer with a chart you can pin to a dashboard.
sequenceDiagram
actor U as You
participant C as clariBI
participant AI as AI engine
participant V as Hugging Face
U->>C: Ask a question about public machine-learning hub data
C->>AI: Plan the analysis
AI->>V: Call the right tool
V-->>AI: Tool result
AI->>AI: Summarize and chart
C-->>U: Answer plus visual
Setting up the connection
- Open Data Sources in the clariBI sidebar.
- Click Add data source.
- Open the MCP Servers tab.
- Click the Hugging Face card.
- Click Authorize with Hugging Face.
- Sign in to Hugging Face in the popup window and grant the requested read scopes.
- Back in clariBI, give your data source a name.
- Click Finish.
Permissions and data access
This is a public, read-only MCP endpoint used anonymously, so no credentials are sent or stored: clariBI sends no login, token, or key. The connector reads public Hub data only (searching models, datasets, Spaces, and papers, reading repository detail, and searching documentation); it cannot create, edit, or delete a repository. Anonymous use is rate-limited by Hugging Face. Disconnect anytime from Settings → Integrations in clariBI.
Troubleshooting
| Error | Cause | Fix |
|---|---|---|
| "Rate limited" or empty results under load | Anonymous access to the Hugging Face MCP server is rate-limited. | Space out heavy queries; for higher limits, Hugging Face documents setting an HF token at hf.co/settings/mcp. |
| "A repository is not found by ID" | The repo ID is not in author/name form, or the repo type filter excludes it. | Pass the full author/name id (for example openai/gpt-oss-120b) and set the repo type when it is not a model. |