Back office ops · Production
Building the Hugging Face MCP Server
The problem
Hugging Face wanted to provide Hub access via MCP with user-customizable tool selection and simple remote access via URL, without requiring complicated downloads or local configuration.
Workflow diagram · grounded in source
1
Client sends initialize request
trigger
“A typical connection sequence is: Initialize, Notifications/Initialize, tools/list and then prompts/list”
2
User credentials lookup
integration
“This is managed using the supplied HF_TOKEN or OAuth credentials that we look up on request”
3
Dynamic tool selection per user
routing
“For anonymous users we supply a standard set of Tools for using the Hub along with an Image Generator. For authenticated users our state comprises their selected tools and chosen Gradio applications”
4
Tool call direct response
output
“Direct Response provides the lowest deployment resource overhead”
Reported outcome
Hugging Face launched hf.co/mcp using Streamable HTTP in a Stateless, Direct Response configuration, attracting 164 different clients in the first week of July 2025, with Gradio Spaces support enabling LLMs to access the latest machine learning applications.
Reported metrics
unique MCP clients in first week164
MCP control messages per tool call100
Reported stack
MCPStreamable HTTPSSESTDIOGradiomcp-remoteOAuthZeroGPUVSCodeCursorclaude.ai
Frequently asked questions
What did this team achieve with this AI workflow?
Hugging Face launched hf.co/mcp using Streamable HTTP in a Stateless, Direct Response configuration, attracting 164 different clients in the first week of July 2025, with Gradio Spaces support enabling LLMs to access…
What tools did this team use?
MCP, Streamable HTTP, SSE, STDIO, Gradio, mcp-remote, OAuth, ZeroGPU, VSCode, Cursor.
What results were reported?
unique MCP clients in first week: 164; MCP control messages per tool call: 100 (source-reported, not independently verified).
How is this back office ops AI workflow structured?
Client sends initialize request → User credentials lookup → Dynamic tool selection per user → Tool call direct response.