tokensaver
Overview
tokensaver is a local MCP server that turns web pages, PDFs, Office files and JSON into compact, LLM-ready text — so your coding agent spends its context on content, not markup.
Ask your agent "read https://go.dev/doc/effective_go with tokensaver" or "fetch this JSON via tokensaver and drop the large lists". tokensaver fetches or opens the thing, strips everything an LLM doesn't need (navigation, scripts, styling, ads, citation marks, running page headers, null fields…) and returns clean Markdown or shrunk JSON — paged, with an outline so the agent can read just the part it needs.
Where to start
- Install it with Homebrew.
- Connect it to your agent — Claude Code, Cursor, Claude Desktop, or any MCP client.
- See Usage for what to do next, with example prompts.
- Read the read and read_json tool references.
- Tune it with Configuration, and check the Security notes before pointing it at untrusted content.
How much it saves
Measured on real sources by the live benchmark (tokens estimated as characters / 4):
| Source | Raw tokens | tokensaver | Saved |
|---|---|---|---|
| Wikipedia article (Go), whole article | 183,202 | 19,383 | 89% |
| …its outline | 248 | 99.9% | |
| GitHub repo page | 87,100 | 1,433 | 98% |
| MDN guide (Using Fetch), whole page | 45,954 | 6,270 | 86% |
| GitHub search API, 50 repos, select + table | 78,745 | 379 | 99.5% |
See the full table and methodology in the README.