Hugging Face for your agents.
Connect Hugging Face once and every agent you run inside Even (Claude Code, Codex, OpenCode, Even Agents) can use it: 9 curated tools, discovered on demand, with your key stored on your machine and every call going direct to Hugging Face.
What agents can do with Hugging Face
Curated for agents: the calls that matter, with schemas and descriptions written for tool use, not an endpoint dump.
huggingface_meReturns the authenticated user's profile.
huggingface_list_modelsSearch models by query string. Sorted by likes descending.
huggingface_get_modelGet details of a specific model by id (e.g. mistralai/Mistral-7B-v0.1).
huggingface_inferenceRun inference on a hosted model. `model` is the model id. `inputs` is the input text or prompt. Uses the free Inference API.
huggingface_list_datasetsSearch datasets on the Hugging Face Hub by query.
huggingface_get_datasetGet details of a specific dataset by its id (e.g. databricks/databricks-dolly-15k).
huggingface_list_spacesSearch Spaces (hosted app demos) on the Hugging Face Hub by query.
huggingface_get_spaceGet details of a specific Space by its id.
huggingface_list_model_tagsList all model tags/categories available on the Hub.
Connect it in a minute
The connect flow is agent-driven: Even opens the key page in its own browser and collects the key through a secure native field. It never passes through the chat.
- Paste the prompt below into any agent running inside Even, or click Connect in Even's catalog.
- Sign in to Hugging Face and go to the Access Tokens page.
- Click 'New token', choose a role (read or write), and name it.
- Copy the token; it is shown once.
- Paste the key into Even's secure field. Even verifies it and your agents are live.
Connect Hugging Face to Even: call connect_begin("huggingface") on the even-connect MCP and walk me through it.
Search for text generation models on Hugging Face Hub and run inference on one.
More Even Connect services
One MCP server, a handful of meta-tools, every service below. Agents search for the right tool at run time, so context stays lean.