Integrations · Development

Connect Hugging Face to Well

Link Hugging Face model, dataset, and author to the customer and revenue context they affect.

What Well pulls from Hugging Face

Hugging Face feeds model, dataset, author into Well as a source. The connection is read-only; disconnect at any time from your workspace settings to revoke Hugging Face’s access.

From Hugging FaceIn Well
Model
Indexed record
Dataset
Workspace context
Author
Team member
Hugging Face logo
Sync mode
MCP hybrid
Refresh
live + reconciliation reads
Direction
Hugging Face → Well

What Well does with your Hugging Face connection

Connect

Connect Hugging Face over OAuth (MCP DCR): Well registers a client through Hugging Face's OAuth Dynamic Client Registration and you approve on Hugging Face's side, so the password never reaches Well; Well holds only a scoped token it can refresh.

Sync

Well brings model, dataset, and author in from Hugging Face on live events backed by periodic reconciliation reads. The first sync backfills history in the background and the connection stays live after.

Enrich

Well resolves each Hugging Face entity into indexed record, workspace context, and team member, assigns categories, and links every record to an audit trail across the rest of your connected tools.

Available

Your data from Hugging Face lands in the workspace as indexed record, workspace context, and team member you can search, chart, and automate.

Questions Hugging Face unlocks

Ask in plain language. Well answers from your connected Hugging Face connection, resolved against the rest of your stack.

Ask Well

Ask anything about your Hugging Face connection

Summarise this month
What needs attention?
Find unmatched records

Ask about your Hugging Face connection…

Connect Hugging Face in three steps

  1. 01

    Authorise Hugging Face via MCP

    From Well's Connections panel, search for Hugging Face and click Connect. Well discovers Hugging Face's MCP OAuth metadata from its .well-known endpoint, registers a client via Dynamic Client Registration, and routes you through the authorize prompt on Hugging Face's side. No client credentials to paste.

    • AuthOAuth (MCP DCR)
    • DiscoveryMCP .well-known
    • MCP serverhuggingface.co/mcp
  2. 02

    Map your entities

    Well's MCP client discovers model, dataset, and author from Hugging Face that Hugging Face's MCP server exposes and brings them into your workspace using the canonical mappings defined in Well's data-views layer. You can inspect the workspace data model from Settings > Data Model.

    • Mappingpreconfigured by Well
  3. 03

    Use the data

    Ask questions in conversation, build records tables, or let agents act on Hugging Face records. Because Hugging Face ships through MCP, Well treats every entity it exposes as queryable graph state alongside the rest of your connected tools.

    • First syncstarts as soon as connect completes

MCP handshake typically completes under a minute on warm connections. Resource enumeration runs immediately after; full backfill of historical Hugging Face records happens in the background.

How Hugging Face stays secure with Well

Well connects to Hugging Face's MCP server (huggingface.co/mcp) via the Model Context Protocol. The MCP server itself handles credential issuance through OAuth Dynamic Client Registration; Well never sees a long-lived Hugging Face secret. Tokens are scoped to the model, dataset, and author Hugging Face exposes and refreshed server-side. Reads only, Well cannot modify or delete Hugging Face records.

OAuth (MCP DCR)Encrypted at restGDPR compliant
Read the full privacy and security policy
Scopes Well requestshuggingface.co/mcp
  • Read Hugging Face records

    Resources the Hugging Face MCP server exposes, scoped by your OAuth approval.

    granted
  • Resolve model, dataset, and author across your stack

    Match identifiers in Hugging Face against the same entities your other connected tools expose, so each record carries cross-tool context.

    granted
  • Modify or delete Hugging Face records

    Not granted; Hugging Face is read-only in Well. Write-back is opt-in per connector when a write surface exists.

    refused
  • Store Hugging Face passwords or session cookies

    Authentication runs through OAuth (MCP DCR) tokens we never see.

    refused
Disconnect any time from Well settings or Hugging Face’s admin panel.

Frequently asked questions about Hugging Face and Well

From Well, open Connections, find Hugging Face, and click Connect. Well discovers the OAuth metadata at huggingface.co/mcp/.well-known/oauth-authorization-server, registers a client on the fly via Dynamic Client Registration, and walks you through the authorize prompt on Hugging Face's side. No client_id or client_secret to paste; the handshake takes roughly a minute.

Well brings in model, dataset, and author from Hugging Face and reshapes each entity into your workspace: model becomes indexed record; dataset becomes workspace context. Every record arrives with the source identifier preserved, so Hugging Face stays the source of truth while Well lets you query the data alongside the rest of your stack.

Every Hugging Face incident, postmortem, or rollback event carries the affected services, which Well joins to the customer entities those services support. Asking "which customers were impacted by last Tuesday's incident" returns the list with severity, restoration time, and any open follow-up actions per customer.

Well watches for updates to models and applies them on the next sync, preserving the original workspace identifier so anything downstream (records tables, conversation history, agent runs) keeps pointing at the right indexed record. Edits flow in incrementally; nothing is duplicated, and the Hugging Face source identifier is retained on every revision.

Well combines live Hugging Face events with periodic reconciliation reads of model, dataset, and author. New rows appear within seconds when Hugging Face pushes them; periodic reads catch anything a push missed. The first connect backfills history from Hugging Face in the background.

Open Connections > Hugging Face in Well and click Disconnect. Well calls Hugging Face's token revocation endpoint and stops calling Hugging Face immediately on its side; provider-side propagation to Hugging Face's admin panel varies by provider. The model, dataset, and author Well already synced stay in your workspace by default; ask Support to purge them if your compliance flow requires it.

Hugging Face records are stored in the region Well operates for your workspace. Specific region details and the steps to change region are available from Support. See /privacy for the current data-handling policy.

Hugging Face logo

Ready to connect Hugging Face?

Connect once. Every model, dataset, author from Hugging Face becomes searchable, queryable, and ready for your agents and tables. Disconnect any time.