MCP server: One MCP server.

ProxyLLM

@proxyllmRun by its publisher: served from their own domain, namespace checked by the registry.MCP server · External

OpenAI-compatible LLM gateway: models, savings math, key provisioning, autonomous email-OTP signup

LLM access & routingv1.0.2MITJavaScript
Freeno charge per call

Runs on your machine from its package; the steps are below.

What it does

  1. Reads
  2. Writes
  3. High impact

At most: moves money, runs code or commands, deletes data, or sends messages for you.

What it costs

Free

Odel charges nothing per call. The publisher may charge for its own service.

Runs locally over stdio

This server isn't hosted — your MCP client launches it from a package registry. Use one of the commands below, or drop the config into your client (e.g. Claude Desktop).

npmproxyllm-mcpv1.0.2

Run

npx -y proxyllm-mcp

MCP client config

{
  "mcpServers": {
    "proxyllm": {
      "command": "npx",
      "args": [
        "-y",
        "proxyllm-mcp"
      ]
    }
  }
}
README2 min read

proxyllm-mcp

MCP server for ProxyLLM, the OpenAI-compatible LLM gateway. It gives any MCP client (Claude Code, Claude Desktop, Cursor, Cline, Windsurf, OpenClaw) live model catalogs, plan-savings math, routing-key introspection and provisioning, and an account signup flow an agent can complete on its own: email OTP in, sk_ account token out, HTTP 402 with a checkout link until a human activates the membership.

Install

Claude Code:

bash
claude mcp add proxyllm -- npx -y proxyllm-mcp

Cursor, Cline, Claude Desktop, or any client that takes a JSON server entry:

json
{
  "mcpServers": {
    "proxyllm": {
      "command": "npx",
      "args": ["-y", "proxyllm-mcp"]
    }
  }
}

No configuration is required to start. Two optional environment variables save passing tokens per call:

  • PROXYLLM_ACCOUNT_TOKEN: an sk_ (read+write) or rk_ (read-only) account token for the management tools.
  • PROXYLLM_ROUTING_KEY: a pllm_ routing key for key introspection and key-scoped model lists.

Tools

Free, no account needed:

  • proxyllm_savings_calculator: monthly OpenAI API bill in, subscription tier + flat total + monthly savings out.
  • proxyllm_list_models: OpenAI-shaped model list from the gateway.
  • proxyllm_introspect_key: what a routing key can do (lanes, models, budget, whether a Codex subscription backs it).

Account flow (an agent can run this end to end):

  • proxyllm_signup: emails a 6-digit code to the operator's inbox. No captcha.
  • proxyllm_verify_signup: exchanges the code for the sk_ account token (returned once).
  • proxyllm_account: account state; while unpaid it carries the checkout link and a message written to be relayed to the operator. Poll until plan is "pro".

Provisioning (paid accounts):

  • proxyllm_create_routing_key: mints a pllm_ key, optionally wiring its provider chain (Codex subscription seats, self-hosted Claude Code bridges, metered API keys) in the same call.
  • proxyllm_list_routing_keys: keys, plus the wireable lane ids with include_lanes.
  • proxyllm_usage: 30-day usage summary.

Unpaid accounts get 402 payment_required on management tools by design; the body names the price ($129/mo flat), the checkout URL, and the 48-hour removal deadline for never-activated accounts. Error bodies are returned verbatim because they contain the next step.

MIT license.

Similar modules

Browse LLM access & routing