Auxen

ai.auxenv1.0.2更新於 Oct 1, 2026

Provision private AI model endpoints on dedicated GPUs (Llama, Qwen, Mistral). Pay per minute.

已驗證Streamable HTTP可網頁執行AI & ML Cloud & Infrastructure Finance

概覽

AI 產生的概覽

讓助理在 Auxen 上開通、查看與銷毀私有專用 GPU 模型端點，依分鐘計費。

功能: 提供工具列出可用模型、列出帳號下的執行個體，以及讀取執行個體的狀態、端點與 api_key。也可讀取美元餘額與生效訂閱、開通新的模型執行個體，以及銷毀既有執行個體。執行個體在專用 GPU 上執行 Llama 3.1、Qwen 2.5、Mistral、Gemma 2、Phi-3 等開源模型。
適用情境: 適合需要助理隨需建立或銷毀私有模型端點，或在操作前查看帳號餘額與執行個體狀態的情境。適用於希望使用專用、非共享推論而非託管 API 的工作流程。若只是呼叫既有模型端點，則不需要安裝。
執行需求: 遠端 StreamableHTTP 端點 OAuth 2.1 + PKCE，或直接使用 Auxen API 金鑰（auxen_live_* / auxen_test_*）以 Authorization: Bearer 傳送。需要 Auxen 帳號並可連線至該端點。

安裝前請注意

開通執行個體會依分鐘計費並產生費用，銷毀執行個體無法復原。伺服器需要憑證：OAuth 2.1 + PKCE 登入，或使用 Auxen API 金鑰（auxen_live_* / auxen_test_*）以 Authorization: Bearer 傳送。執行個體狀態回應中包含 api_key，應將工具輸出視為敏感資料。

安裝

在 SourceWeft 中

開啟儀表板中的 Auxen，將其新增到工作區。
為需要使用其工具的對話啟用該服務。

Web executable，透過 Streamable HTTP。遠端服務在工作區中設定後即可從網頁執行環境執行。

其他 MCP 客戶端

把它新增到你客戶端的 mcpServers 設定中。

{
  "mcpServers": {
    "auxen": {
      "type": "http",
      "url": "https://api.auxen.ai/mcp"
    }
  }
}

README

Auxen MCP Server

This is the public manifest repo for the Auxen MCP (Model Context Protocol) server. The server itself runs at https://api.auxen.ai/mcp — this repo exists so registries (Smithery, Glama, the official MCP registry) have a canonical place to read metadata from.

Connect

The Auxen MCP server is a remote, StreamableHTTP server. Add it to your MCP client by URL:

Authentication uses OAuth 2.1 + PKCE (recommended for browser-based clients) or a direct Auxen API key (auxen_live_* / auxen_test_*) sent as Authorization: Bearer <key>.

OAuth flow

The discovery metadata is at:

https://api.auxen.ai/.well-known/oauth-authorization-server (RFC 8414)
https://api.auxen.ai/.well-known/oauth-protected-resource (RFC 9728)

Clients that support Dynamic Client Registration (RFC 7591) — including Claude.ai's Connectors Directory — can register themselves automatically. After registration the client redirects the user's browser to https://api.auxen.ai/oauth/authorize, the user logs in to Auxen and approves the connection on https://auxen.ai/oauth/authorize, and the client receives an authorization code that exchanges for an access token at https://api.auxen.ai/oauth/token.

Direct key (programmatic)

For agents that don't go through a browser, generate an auxen_live_* (or auxen_test_*) key at https://auxen.ai/dashboard/api-keys and send it as Authorization: Bearer <key> on every MCP call.

Tools

Tool	Effect	Hint
`auxen_list_models`	List available models, optionally filtered by size	read-only
`auxen_get_instance_status`	Get status, endpoint, api_key for an instance	read-only
`auxen_list_instances`	List all instances on the account	read-only
`auxen_get_balance`	Read USD credits + active subscriptions	read-only
`auxen_provision_model`	Provision a new model instance — spends money	destructive
`auxen_destroy_instance`	Destroy an instance — irreversible	destructive

What is Auxen?

Auxen provisions private, dedicated GPU instances running open-source models (Llama 3.1, Qwen 2.5, Mistral, Gemma 2, Phi-3). Each instance is fully private — no shared inference, no third-party routing. Pay-per-minute billing, no subscriptions.

For the human-facing documentation:

Website: https://auxen.ai
Architecture explainer: https://auxen.ai/architecture
Agent reference: https://auxen.ai/agent-reference.md
Concise reference for LLMs: https://auxen.ai/llms.txt
OpenAPI 3.1 spec: https://api.auxen.ai/openapi.json

Support

Email: [email protected]
Status: https://status.auxen.ai

來源：README.md，提交 02a6479

工具

工具後設資料尚未被收錄。

版本歷史

v1.0.2最新Sep 16, 2026