Meet Token Saver: Open-Source MCP Extension Cuts Claude PDF Token Costs 90-99%
Marktechpost AI releases Token Saver, an MCP extension using Local Hybrid RAG to query massive PDFs without uploading files.
Stock photo for illustration only, not from the actual event
- Developed by RIT student Arnav Rai, supervised by Jean-marc Mommessin and Asif Razzaq
- Cuts Claude token consumption by an impressive 92% to 99% instantly
- Operates entirely via Local Hybrid RAG on your machine without Python environments
- Guarantees strict data privacy as documents never leave your local hard drive
The Marktechpost AI team has rolled out Token Saver, an open-source Model Context Protocol (MCP) extension designed for Claude Desktop (MIT licensed, currently at v1.0). Developed at Marktechpost AI Media Inc by Arnav Rai, a computer science undergraduate at the Rochester Institute of Technology during his internship, with supervision from Jean-marc Mommessin and Asif Razzaq, Token Saver fundamentally shifts how Claude engages with local documents. By implementing a Local Hybrid RAG system directly on your local machine, it enables users to query massive PDF files without ever uploading the actual file to the AI model.
Token consumption is drastically slashed by 92% to 99%, privacy is completely guaranteed, and the setup requires exactly zero Python environments or terminal configurations. Most users assume that dropping a PDF into Claude merely extracts text. In reality, Claude's default behavior converts each page into an image to preserve charts and layouts while simultaneously extracting text. Before a single image token is even counted, the raw text alone can consume 1,500 to 3,000 tokens per page.
While Prompt Caching and Claude Projects help soften this blow, they fail to resolve the core issue: the entire document still crosses the boundary to the provider's servers. Furthermore, if an analyst only requires two relevant paragraphs from a 1,000-page textbook, forcing the LLM to search through all 1,000 pages independently is both inefficient and highly susceptible to hallucination.
Stock photo for illustration only, not from the actual event
Marktechpost's Token Saver operates as a local MCP server—a lightweight background utility on your machine that Claude can call as a tool, ensuring the PDF never leaves your hard drive. When you submit a query, Token Saver leverages Local Hybrid RAG to pinpoint the answer. Hybrid RAG stands as the gold standard for document retrieval because it synergizes two potent search mechanisms: keyword matching and semantic search.
Deploying Retrieval-Augmented Generation (RAG) locally represents a crucial paradigm shift as enterprise data privacy concerns escalate. By filtering out irrelevant context and supplying only targeted text slices to the LLM, this approach drastically curtails API expenses while maximizing context window efficiency, resulting in higher accuracy and fewer hallucinations.
By harmonizing these two approaches locally, the server scans the file and feeds Claude only the highly relevant passages. Claude then synthesizes the response, citing exact page numbers and providing a running tally of the tokens just saved. The extension operates entirely within a single, long-running local process, executing eight rapid steps through the Local Hybrid RAG pipeline before any text reaches Claude (note that the embedding model is optional, falling back gracefully to keyword-only matching if loading fails).
"Token Saver fundamentally changes how Claude interacts with local documents by implementing a Local Hybrid RAG system directly on your machine."
Marktechpost AI Team
Source: MarkTechPost
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment