Auxen MCP Server

Auxen's hosted Model Context Protocol server. Lets Claude, Cursor, Continue, or any MCP client provision, manage, and tear down private AI model instances on dedicated GPUs — using natural language.

Endpoint:        https://api.auxen.ai/mcp
Transport:       Streamable HTTP (stateless)
Auth:            OAuth 2.1 + PKCE-S256 + Dynamic Client Registration (RFC 7591)
Discovery:       /.well-known/oauth-authorization-server (RFC 8414)
                 /.well-known/oauth-protected-resource (RFC 9728)
Spec version:    MCP 2025-06-18

Quickstart

Add to Claude.ai

1. In Claude.ai → Settings → Connectors → Add custom connector
2. URL: https://api.auxen.ai/mcp
3. Click Connect → walk through the OAuth consent → Approve

That's it. Claude can now call all 10 Auxen tools.

Add to Cursor / Continue / other MCP clients

In your client's MCP config:

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

The client will discover OAuth automatically and prompt for browser consent on first use.

Tools

10 tools across two permission groups (read-only / write-or-destructive). Every destructive tool requires explicit user approval through the client's standard tool-call UX.

Read-only

Tool	What it does
auxen_list_models	List available AI models with size, parameter count, and PAYG rate
auxen_get_instance_status	Get an instance's status, endpoint URL, API key, hourly burn
auxen_list_instances	List all instances on the account (active and historical)
auxen_get_balance	USD credit balance, active instance count, current hourly burn
auxen_get_schedule	Read an instance's schedule (always_on, as_needed, scheduled_window)

Write / destructive

Tool	What it does
auxen_provision_model	Spin up a new private model instance. Draws from PAYG balance.
auxen_destroy_instance	Tear down an instance, invalidate endpoint and API key. Irreversible.
auxen_pause_instance	Pause a running instance — compute meter stops; minimal idle charges.
auxen_wake_instance	Wake a paused instance. ~3-minute provisioning window.
auxen_set_schedule	Configure when an instance is allowed to run.

What is Auxen?

Auxen hosts private AI model endpoints on dedicated GPUs. Pay by the minute, no subscriptions, OpenAI-compatible API.

SIZE     MODELS                              RATE (1× capacity)
─────────────────────────────────────────────────────────────────
Small    gemma2-2b, mistral-7b, llama3.2-3b, phi3-mini    $0.10/hr
Medium   llama3.1-8b, qwen2.5-14b, mistral-nemo-12b, …    $0.20/hr
Large    qwen2.5-32b, mistral-small-24b, command-r, …     $0.65/hr
XL       llama3.1-70b, qwen2.5-72b, mixtral-8x22b          $1.50/hr

Each Auxen instance is fully private — no shared inference, no per-token fees, no multi-tenant routing.

Verify the server is live

$ curl -i https://api.auxen.ai/.well-known/oauth-protected-resource
HTTP/1.1 200 OK
content-type: application/json

{
  "resource": "https://api.auxen.ai/mcp",
  "authorization_servers": ["https://api.auxen.ai"],
  "scopes_supported": ["mcp"],
  "bearer_methods_supported": ["header"]
}

$ curl -i -X POST https://api.auxen.ai/mcp -H 'Content-Type: application/json' -d '{}'
HTTP/1.1 401 Unauthorized
www-authenticate: Bearer realm="https://api.auxen.ai/mcp",
                  resource_metadata="https://api.auxen.ai/.well-known/oauth-protected-resource/mcp"

Related

• Auxen homepage
• Auxen docs
• Auxen Python SDK — pip install auxen
• Auxen AI SDK Provider — pnpm add @auxen-ai/ai-sdk-provider
• MCP specification

License

Apache-2.0