Model Context Protocol

Let your agent search Inferix directly

An MCP server that gives Claude, Cursor or any MCP-compatible client five tools for finding models, datasets and rentable GPUs on Inferix — so the agent looks things up itself instead of asking you to paste them in.

It runs locally over stdio. Nothing is hosted, nothing is proxied, and read-only calls work without a token.

npx -y inferix-mcp-serverGet an API token

The five tools

This is the whole surface. Each one is read-only — the server can look things up on Inferix, and cannot create, change or delete anything in your account.

inferix_search_models

Search the model hub by free-text query, task or library.

query · task · limit

inferix_get_model

Full detail for one model by namespace and name — card, files, tags and licence.

namespace · name

inferix_search_datasets

Search datasets by free-text query.

query · limit

inferix_search_gpus

Search the GPU marketplace for rentable machines, filtered by GPU type and maximum hourly price.

gpu · maxPrice · limit

inferix_market_metrics

Current marketplace supply, demand and pricing trends.

no arguments

Adding it to a client

Drop this into your MCP client's config — claude_desktop_config.json for Claude Desktop, or the equivalent for Cursor and other clients. Restart the client and the five tools appear.

{
  "mcpServers": {
    "inferix": {
      "command": "npx",
      "args": ["-y", "inferix-mcp-server"],
      "env": {
        "INFERIX_API_URL": "https://inferix.co/api",
        "INFERIX_API_TOKEN": "hf_your_token_here"
      }
    }
  }
}

INFERIX_API_URL

Defaults to https://inferix.co/api. Point it at your own deployment if you self-host.

INFERIX_API_TOKEN

Optional. Public models, datasets and GPU listings are readable without one; a token adds anything private to your account.

What it is useful for

Picking a model without leaving the editor. Ask for a small multilingual embedding model and the agent searches the hub and reads the cards itself.
Costing a job before running it. The GPU search and market-metrics tools let an agent compare what is actually available and at what hourly price.
Keeping context small. The agent fetches the one model card it needs rather than you pasting several.

Source lives in mcp-server/ in the Inferix repository — five tools, about a hundred lines, no hidden calls.