Make your Claude and ChatGPT usage go further.
TokenBooster lets Claude, Claude Code, ChatGPT or Codex hand bulk work to cheaper open models on your own API keys. That covers summarizing long files, extracting data, classifying rows, reformatting and drafting boilerplate. Your main model stays in charge: it writes the brief, checks the result and answers you.
Every result passes quality checks (schema, verbatim evidence, counts, secrets, injected instructions). If a check fails twice, or delegating wouldn't save anything, your main model just does the task itself.
How it works
- Your model writes a short brief: goal, rules, output schema.
- TokenBooster fetches the input itself (from a URL or an upload), so your model never reads the bulk.
- A cheaper model does the work. TokenBooster gates the result.
- Your model gets a compact, checked result, or a handback telling it to do the task directly.
Honest numbers only: savings shown in your stats are labelled estimates. Worker usage and spend come from your provider's reports.
1. Create an account
2. Add a worker key
Any OpenAI-compatible endpoint works (most open-model providers offer one). Your key is encrypted and never shown again except its last 4 characters.
3. Connect your client
MCP endpoint (streamable HTTP), with your token as a bearer header:
URL: https://tokenbooster.ai/mcp Header: Authorization: Bearer tb_YOUR_TOKEN
Claude Code:
claude mcp add --transport http tokenbooster https://tokenbooster.ai/mcp --header "Authorization: Bearer tb_YOUR_TOKEN" export TOKENBOOSTER_TOKEN=tb_YOUR_TOKEN # used for file uploads
Other clients: add a remote MCP server with the URL and header above. Check your client's docs for where to set headers.