MCP Gateway · Claude Cowork
Connect Hugging Face to Claude Cowork
Comparing models for a use case takes hours of searching, reading model cards and writing up. Cowork can run that research in one long session once Hugging Face arrives through a plugin in the organization's synced marketplace. Agent hooks pass each prompt to Speakeasy for a decision first, and the OTLP export records who ran it.
- Category
- AI & search
- Tools
- 10
- Authentication
- OAuth
- Claude Cowork
- Cowork tab in Claude Desktop
Behind the MCP Gateway
Run Hugging Face MCP behind the MCP Gateway
One URL for every agent, sign-in through your IdP, RBAC down to the tool, and every Hugging Face call inspected and logged.
Set up
Add Hugging Face to Claude Cowork in 4 steps
- 01
Add Hugging Face from the catalog
An admin finds Hugging Face in the MCP Catalog (MCP Gateway > MCP > Catalog) and clicks Add. Speakeasy registers the vendor's server as a remote MCP server and proxies sessions to it. Nothing is copied or re-hosted.
Hugging Face uses OAuth. Servers that support dynamic client registration get their OAuth configuration applied automatically.
Where MCP-scoped policies are enabled, the install offers a recommended risk policy for the new server that detects secrets and PII and inspects tool requests and responses.
- 02
Choose who can use Hugging Face
Give teams the
mcp:connectscope on the Hugging Face server through team access or a role. Claude Cowork connects to the Hugging Face server's own URL, and the server's authorization decides who can connect and which tools they can call. - 03
Connect Claude Cowork
Sync the organization's plugin marketplace into Claude, then make the plugins available in Cowork.
New MCP servers join the project's default plugin automatically. Each server's overview page shows which plugin it is published to, with Publish or Update to change it.
Add the importing admin's GitHub username as a marketplace collaborator and accept the invite. Only that admin needs it.
In Claude Organization settings > Plugins & skills, select Add > Sync from GitHub and pick the organization's plugin repository. The Claude GitHub App must be installed on it.
Mark the observability plugin Required, and set the plugin that carries this server to Available to install or Installed by default.
In the Cowork tab's Monitoring settings, set the OTLP endpoint, protocol and headers.
- 04
Sign in and use Hugging Face from Claude Cowork
The first time Claude Cowork connects, it opens a browser so the person can sign in to Speakeasy through your identity provider.
Hugging Face needs its own OAuth grant, so each person connects it once. Until they do, Hugging Face calls fail while the rest of the gateway keeps working.
From then on, Hugging Face's tools appear in Claude Cowork, limited to what the person's access allows, and every call lands in the gateway's tool logs.
Cowork reads the synced repository, so a new session picks up plugin changes. Members who receive plugins through Cowork don't need a GitHub invitation.
Use cases
What Claude Cowork users do with Hugging Face
Write a model landscape report
Cowork finds the most liked models for a task, reads each one's repository details, and writes a comparison report your team can use to choose a starting point.
Pair models with their papers
For each shortlisted model, Cowork searches for the research paper behind it and adds a short summary and link to the report.
Process a folder of scans
Give Cowork a batch of scanned forms. It runs an OCR Space task on each one and assembles the extracted text into a single document.
Tools
10 tools
Each Hugging Face tool can be allowed or denied per team and role at the MCP Gateway before Claude Cowork sees it.
hf_whoami
Hugging Face tools are being used by the authenticated user
space_search
Find Hugging Face Spaces using semantic search. IMPORTANT Only MCP Servers can be used with the dynamic_space toolInclude links to the Space when presenting the results.
hub_repo_search
Search Hugging Face repositories with a shared query interface. You can target models, datasets, spaces, or aggregate across multiple repo types in one call. Use space_search for semantic-first discovery of Spaces. Include links to repositories in your response.
paper_search
Find Machine Learning research papers on the Hugging Face hub. Include 'Link to paper' When presenting the results. Consider whether tabulating results matches user intent.
hub_repo_details
Get details for one or more Hugging Face repos (model, dataset, or space). Auto-detects type unless specified.
hf_doc_search
Search and Discover Hugging Face Product and Library documentation. Send an empty query to discover structure and navigation instructions. Knowledge up-to-date as at 8 April 2026. Combine with the Product filter to focus results.
Control
What Speakeasy can see and stop in Claude Cowork
The gateway covers every client the same way. The rest depends on what Claude Cowork exposes to Speakeasy.
Tool logs
Gateway
Every Hugging Face call is logged with the person and the client that made it.
Access by role
Gateway
People reach only the Hugging Face tools their team or role allows.
Request and response blocking
Gateway
A risk policy scoped to Hugging Face can reject a tool request before it reaches the server, or withhold a result from Claude Cowork. MCP-scoped policies are enabled per organization.
Prompt blocking
Claude Cowork
Agent hooks send each prompt to Speakeasy, which allows or denies it before the step runs.
Session telemetry
Claude Cowork
Cowork's OTLP export ties sessions to people and feeds observability.
Anthropic inference hooks
Claude Cowork
On Claude Enterprise, prompts can be checked before inference (beta).
- Neither the device agent nor MDM reaches Cowork, so the Claude-side setup always applies.
- Personal Claude accounts can't be governed in Cowork.
Questions
Frequently asked questions
How does Hugging Face reach Cowork?
Can Cowork use Hugging Face from a personal Claude account?
How do Claude Cowork users get Hugging Face?
Are Hugging Face calls from Claude Cowork logged?
Connect more