token-save-mcp

# token-save-mcp **Your coding agent burns its context reading files. This stops it — and shows you the receipt.** [](https://github.com/Habartru/token_save_mcp/actions/workflows/ci.yml) [](https://pypi.org/project/token-save-mcp/) [](https://github.com/Habartru/token_save_mcp/blob/main/LICENSE) [](https://www.python.org/downloads/) Read one 4,000-line file and ~35,000 tokens sit in your agent's context for the rest of the session. Read a few more and it compacts, forgets your instructions, and every later turn costs more — because each one pays for the whole accumulated context again. This MCP server sends those files to a cheap worker model instead and returns only the answer. The bytes are paid for once, in the worker's context, not permanently in yours.  The hook **blocks** the expensive read and redirects it. The answer comes back with the worker's **real token usage from the API response** — not an estimate, a receipt: ``` ───────────────────────────────────────────────────────────── token-save: 1 file, 606 lines | direct read ≈7,042 tok → into context ≈234 tok (saved 6,808 · 97%) worker: glm-5.3-flash | 6,155 in / 278 out | 4.0s ``` *(`glm-5.3-flash` is just the worker configured in that run — you pick your own.)* --- ## Install Two commands, plus an API key of your own. ```bash pip install token-save-mcp token-save-mcp init --hook ``` `init` finds a provider key you already have, registers the MCP server with your agent, and installs the hook. If you have no key yet it prints the options and where to get one. One package, one command. There is no second MCP server to ad…