Auxen MCP Server for Claude

Overview

Auxen.ai provides a hosted Model Context Protocol server that lets Claude and other MCP clients manage GPU inference workloads. After OAuth 2.1 + PKCE authorisation or Bearer token setup, assistants can list available models and running instances, check account balance, provision new models, and destroy instances without leaving chat.

The public manifest lives in the auxen-ai/auxen-mcp repository. The server is hosted at https://api.auxen.ai/mcp.

Features

• List available GPU models, running instances, and account balance.
• Provision new inference instances with provision_model.
• Tear down instances with destroy_instance.
• OAuth 2.1 + PKCE or Bearer token auth with auxen_live_* and auxen_test_* keys.
• Hosted remote HTTP transport compatible with Claude Connectors and Claude Code.

Use Cases

• Check GPU availability and pricing before launching a fine-tuning job.
• List running instances and idle spend during a cost review.
• Provision a test instance with an auxen_test_* key before promoting to production.
• Destroy orphaned instances after an experiment completes.
• Ask natural-language questions about current balance and active deployments.

Installation

Claude (Connectors)

1. Add a custom connector under Settings → Connectors.
2. Enter the MCP URL: https://api.auxen.ai/mcp.
3. Complete OAuth 2.1 + PKCE sign-in, or configure a Bearer token with your API key.
4. Enable the connector in your next conversation.

Claude Code

claude mcp add --transport http auxen https://api.auxen.ai/mcp
claude mcp list

For API key auth, add the Authorization: Bearer auxen_live_YOUR_KEY header in your MCP config.

Other MCP clients

Add a custom remote connector pointing at https://api.auxen.ai/mcp and complete OAuth or Bearer token setup supported by your client.

Configuration

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

Omit the headers block when using OAuth through a connector UI. Use auxen_test_* keys for non-production environments.

Examples

List available models

List the GPU models available on my Auxen account and summariseVRAM and pricing tiers.

Check balance and instances

What is my Auxen account balance and how many instances are currently running?

Provision a test instance

Provision a small test inference instance using the cheapest available model in my Auxen test environment.

Security

• Treat provision_model and destroy_instance as destructive; review tool arguments before approving.
• Revoke OAuth tokens and rotate API keys when offboarding users or connectors.
• Use auxen_test_* keys in development to avoid accidental production charges.
• Instance metadata may contain deployment details; avoid pasting raw output into public channels.

Troubleshooting

OAuth or 401 errors

Re-authorise the connector and confirm your Auxen account is active. For Bearer auth, verify the key prefix matches your environment (auxen_live_* vs auxen_test_*).

Provision failures

Check account balance and model availability in the Auxen dashboard. Some models require elevated quotas or billing setup.

Empty instance list

Confirm you are authenticated against the correct Auxen environment. Test keys only see test resources.

Unexpected charges

Review recent provision_model tool calls in your MCP client audit log and destroy unused instances promptly.