ScaledThought README

ScaledThought MCP server

ScaledThought is an OpenAI-compatible API for open-weight models, set up by your coding agent. This repository holds the public metadata for its remote MCP server: this README, the MCP Registry entry (server.json, name com.scaledthought/mcp) and a security policy. It contains no application source code.

OpenAI-compatible API for open-weight models. Any OpenAI SDK works with the base URL https://api.us.scaledthought.com/v1.
Your agent signs up from one pasted prompt. It creates the account and the API key itself.
Prepaid credits, no subscription. The agent asks you to approve a top-up through an approval link; you pay only for what you use.
Budgets per key.
Prompts and completions are not stored or logged. Every response carries x-st-data-retention: none.
US region live; Canada coming soon.

The models live in the US right now and their prices: https://scaledthought.com/pricing, or npx -y scaledthought@latest models --live.

1. Get an API key

Paste this into your coding agent (Claude Code, Codex, Cursor, Cline, OpenCode and others):

text

Set me up on ScaledThought: scaledthought.com/go/org-ghrepo

Or run the CLI yourself:

Terminal

npx -y scaledthought@latest init --region us --source org-ghrepo

init writes ST_API_KEY and ST_BASE_URL to ./.env. Usage is prepaid: before the first call, the agent prints an approval link for you to add credits by card.

2. Connect the MCP server

Remote server, streamable HTTP, authenticated with your API key:

URL: https://api.us.scaledthought.com/mcp?source=org-ghrepo
Header: Authorization: Bearer <your ScaledThought API key>

Claude Code (run it in the directory where init wrote ./.env):

bash

set -a; . ./.env; set +a
claude mcp add --transport http scaledthought 'https://api.us.scaledthought.com/mcp?source=org-ghrepo' \
  --header "Authorization: Bearer $ST_API_KEY"

A project .mcp.json (Claude Code fills in ${ST_API_KEY} from the environment: export it from ./.env before starting Claude Code):

config.json

{
  "mcpServers": {
    "scaledthought": {
      "type": "http",
      "url": "https://api.us.scaledthought.com/mcp?source=org-ghrepo",
      "headers": { "Authorization": "Bearer ${ST_API_KEY}" }
    }
  }
}

Cline (cline_mcp_settings.json; paste the ST_API_KEY value from ./.env in place of <your key>):

config.json

{
  "mcpServers": {
    "scaledthought": {
      "type": "streamableHttp",
      "url": "https://api.us.scaledthought.com/mcp?source=org-ghrepo",
      "headers": { "Authorization": "Bearer <your key>" }
    }
  }
}

What the server does

It manages your ScaledThought account from the agent: account status and settings, API keys, budgets, usage, prepaid top-ups (approval links), webhooks, model search and docs search. Inference itself goes to the OpenAI-compatible API at https://api.us.scaledthought.com/v1.

License

The text and metadata in this repository are MIT licensed (LICENSE). The ScaledThought name and logo are trademarks of ScaledThought Inc. and are not licensed.

ScaledThought MCP server

1. Get an API key

2. Connect the MCP server

What the server does

Links

License