Skip to main content

LLM gateway

Route your team's AI model calls through MintMCP's hosted LLM proxy. Both local agents (Claude Code and other AI coding tools running on developer machines) and MintMCP Coworker Agents can connect to the gateway, so your organization sees all model usage in one place.

LLM gateway home page showing the endpoint and connected agents

The LLM gateway is in early access. Contact us to enable it for your organization.

What you get​

  • Central credentials: agents authenticate with the gateway endpoint using a gateway key. Individual developers don't need their own model API keys.
  • Governance: apply rules to LLM traffic the same way you apply them to MCP tool calls.

Connect a local agent​

Configure a local agent (for example, Claude Code) to route its model calls through the gateway.

Agent configuration form showing gateway endpoint and key fields
  1. In MintMCP, go to LLM Gateway.
  2. Copy the Gateway endpoint URL and your agent key from the setup page.
  3. In your agent's configuration, set the API base URL to the gateway endpoint and the API key to your agent key.

For Claude Code, set these in ~/.claude/settings.json:

{
  "env": {
    "ANTHROPIC_BASE_URL": "<gateway-endpoint-url>",
    "ANTHROPIC_API_KEY": "<your-agent-key>"
  }
}

Connect a Coworker Agent​

To route a Coworker Agent's model calls through the gateway, open the agent's Settings tab and configure the harness:

  1. Under Harness, set Provider to Claude-compatible API.
  2. Enter the Gateway endpoint URL as the API base URL.
  3. Add your agent key as a secret, then reference it in the API key field.

See Harness and model for the full harness configuration reference.

Next steps​

  • Access control: issue and revoke gateway keys, manage per-agent credentials
  • Policy and audit: model governance, rules, usage limits, and audit logging