Token Saver：用本地混合 RAG 将 Claude PDF token 消耗削减 92%-99% 的开源 MCP 扩展

AI 开发者、研究人员和专业人士在使用大型语言模型（LLM）分析大型文档时，经常会遇到一个令人沮丧的障碍：上下文窗口的隐藏、累积成本。将一份 200 页的 PDF 粘贴到聊天中并不是一次性收费。因为每次对话轮次都会将对话历史重新发送给模型，所以每次后续提问时，这份庞大的文档都会再次产生费用。

Marktechpost AI 团队刚刚发布了 Token Saver：https://github.com/Marktechpost/Token-Saver，这是一个开源的 Claude Desktop 模型上下文协议（MCP）扩展（MIT 许可，目前版本为 v1.0）。它由 Arnav Rai：https://www.linkedin.com/in/arnav-rai-475033243/（罗切斯特理工学院计算机科学学生）在 Marktechpost 实习期间开发，由 Jean-marc Mommessin：https://www.linkedin.com/in/contactjmm/ 和 Asif Razzaq：https://www.linkedin.com/in/asifrazzaq/ 监督。Token Saver 从根本上改变了 Claude 与本地文档的交互方式。通过实现本地混合 RAG：https://github.com/Marktechpost/Token-Saver（在你自己的机器上运行系统），它允许你在不将实际文件上传到模型的情况下，向大型 PDF 提问。

大多数用户以为他们将 PDF 放入 Claude 时，它只是简单地提取文本。实际上，Claude 的默认行为是将每一页转换为图像以保留图表和布局，同时单独提取文本。在计数任何图像 token 之前，仅文本就可能每页就达到 1,500 到 3,000 个 token。

虽然提示缓存（Prompt Caching）和 Claude Projects 能帮助缓解这一影响，但它们并不能解决核心问题：整个文档仍然需要传输到提供商的服务器。此外，如果你只需要一份 1,000 页教材中的两段相关内容，强制 LLM 自行搜索整整 1,000 页既低效又容易产生幻觉。

Marktechpost 的 Token Saver：https://github.com/Marktechpost/Token-Saver 相当于一个本地 MCP 服务器：在你的机器上运行的轻量级后台程序，Claude 可以把它当作一个工具调用。PDF 永远不会离开你的硬盘。

当你提问时，Token Saver 利用本地混合 RAG 查找答案。混合 RAG 是文档检索的黄金标准，因为它结合了两种强大的搜索方法：

通过在本地将这两种方法结合，服务器会搜索文件并只将高度相关的内容呈现给Claude。然后Claude综合回答，同时引用确切的页码，并提供你刚刚节省的令牌数量的实时统计。

Token Saver：https://github.com/Marktechpost/Token-Saver 完全在单个长期运行的本地进程中运行。在任何文本到达Claude之前，本地混合RAG管道会执行八个快速步骤：

（注：嵌入模型是可选的，但推荐使用。如果加载失败，系统会优雅地回退到仅关键词匹配）。

由于混合RAG管道会限制返回的文本片段，令牌节省会随着文档大小呈指数增长。以下是Token Saver在基准测试中的表现（通过tiktoken测量，假设每个文档一个问题）：

免责声明：令牌计数使用cl100k_base作为代表，基线假设每次搜索都会计入整个文档。

对于反复处理法院判决、技术标准、财务报告或密集教科书的专业人士来说，Token Saver消除了重复的令牌消耗。

Marktechpost的Token Saver适用于所有三个Claude 3.5等级，但测试显示其表现存在差异：

与许多需要复杂Python环境和JSON配置的开源AI工具不同，Token Saver打包为单个.mcpb文件。

在选择文件夹之前，扩展程序不会运行，因为这个选择同时有三个作用。

这个文件夹值得稍作思考。它设置了可读取内容的边界，使你可以通过名称而不是路径请求文件，并且是服务器在对话开始时显示给Claude的列表。一个只包含你打算查询内容的小文件夹效果要远好于指向整个Documents。

查看GitHub仓库：https://github.com/Marktechpost/Token-Saver。

参考资料：MCP Bundle 格式：https://github.com/anthropics/mcpb | all-MiniLM-L6-v2：https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2 | pdfkb-mcp：https://github.com/juanqui/pdfkb-mcp | wesleygriffin/pdfrag：https://github.com/wesleygriffin/pdfrag | 语料库：Dobbs 诉 Jackson Women’s Health Organization；伯克希尔·哈撒韦 2023 年年度报告

Arnav 目前是罗切斯特理工学院的学生，正在攻读计算机科学学士学位并辅修经济学，拥有实践型后台开发经验，同时是 Marktechpost 的撰稿人，在该平台上撰写关于人工智能/机器学习研究的文章。

Jean-marc 是一位成功的人工智能商业高管。他领导并加速人工智能解决方案的增长，并于 2006 年创办了一家计算机视觉公司。他是人工智能会议上的知名演讲者，并获得斯坦福大学的工商管理硕士学位。

Asif Razzaq 是 Marktechpost Media Inc. 的首席执行官。作为一名有远见的企业家和工程师，Asif 致力于利用人工智能的潜力造福社会。他最近的项目是推出人工智能媒体平台 Marktechpost，该平台以深入报道机器学习和深度学习新闻而著称，这些内容既技术可靠，又易于广大受众理解。该平台月浏览量超过 200 万次，显示了其在观众中的受欢迎程度。

AI developers, researchers, and professionals frequently hit a frustrating wall when analyzing large documents with LLMs: the hidden, compounding cost of context windows. Pasting a 200-page PDF into a chat isn’t a one-time charge. Because the conversation history is re-sent to the model on every single turn, that massive document is paid for again with every follow-up question.

Marktechpost AI team just released Token Saver：https://github.com/Marktechpost/Token-Saver , an open-source Model Context Protocol (MCP) extension for Claude Desktop (MIT licensed, currently at v1.0). It was developed at Marktechpost AI Media Inc by Arnav Rai：https://www.linkedin.com/in/arnav-rai-475033243/ (CS student at Rochester Institute of Technology) during his internship at Marktechpost, supervised by Jean-marc Mommessin：https://www.linkedin.com/in/contactjmm/ and Asif Razzaq：https://www.linkedin.com/in/asifrazzaq/. Token Saver fundamentally changes how Claude interacts with local documents. By implementing a Local Hybrid RAG ：https://github.com/Marktechpost/Token-Saver (system directly on your machine, it allows you to ask questions about massive PDFs without ever uploading the actual file to the model.

Most users assume that when they drop a PDF into Claude, it simply extracts the text. In reality, Claude’s default behavior converts each page into an image to preserve charts and layouts, while separately extracting text. Before a single image token is even counted, the text alone can run 1,500 to 3,000 tokens per page.

While Prompt Caching and Claude Projects help soften this blow, they don’t solve the core issue: the entire document still crosses the boundary to the provider’s servers. Furthermore, if you only need two relevant paragraphs from a 1,000-page textbook, forcing the LLM to search the entire 1,000 pages itself is both inefficient and prone to hallucination.

Marktechpost’s Token Saver ：https://github.com/Marktechpost/Token-Saver acts as a local MCP server : a lightweight background program on your machine that Claude can call as a tool. The PDF never leaves your hard drive.

When you ask a question, Token Saver utilizes Local Hybrid RAG to find the answer. Hybrid RAG is the gold standard for document retrieval because it combines two powerful search methods:

By blending these two approaches locally, the server searches the file and hands Claude only the highly relevant passages. Claude then synthesizes the answer, citing the exact page numbers and providing a running tally of the tokens you just saved.

Token Saver：https://github.com/Marktechpost/Token-Saver runs entirely inside a single, long-running local process. Before any text reaches Claude, the local Hybrid RAG pipeline executes eight rapid steps:

(Note: The embedding model is optional but recommended. If it fails to load, the system gracefully falls back to keyword-only matching).

Because the Hybrid RAG pipeline caps the returned text slice, the token savings grow exponentially with the size of the document. Here is how Token Saver performed in benchmarking (measured via tiktoken, assuming one question per document):

Disclaimer: Token counts use cl100k_base as a stand-in, and baselines assume the whole document is charged on every search.

For professionals working repeatedly with court opinions, technical standards, financial reports, or dense textbooks, Token Saver eliminates the repetitive token tax.

Marktechpost’s Token Saver works across all three Claude 3.5 tiers, but testing revealed distinct behaviors:

Unlike many open-source AI tools that require complex Python environments and JSON configurations, Token Saver is packaged as a single .mcpb bundle.

The extension will not run until a folder is chosen, because that single choice serves three purposes at once.

That folder is worth a moment’s thought. It sets the boundary of what can be read, it is what lets you ask for a file by name instead of typing a path, and it is the list the server shows Claude when a conversation starts. A small folder holding only what you intend to ask about works considerably better than pointing it at all of Documents.

Check out the GitHub Repo：https://github.com/Marktechpost/Token-Saver.

References: MCP Bundle format：https://github.com/anthropics/mcpb | all-MiniLM-L6-v2：https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2 | pdfkb-mcp：https://github.com/juanqui/pdfkb-mcp | wesleygriffin/pdfrag：https://github.com/wesleygriffin/pdfrag | Corpora: Dobbs v. Jackson Women’s Health Organization ; Berkshire Hathaway 2023 Annual Report

Arnav is currently a student at Rochester Institute of Technology pursuing a Bachelor’s degree in Computer Science and a minor in Economics with hands-on backend development experience, and is a contributor at Marktechpost, where he writes about AI/ML research.

Jean-marc is a successful AI business executive .He leads and accelerates growth for AI powered solutions and started a computer vision company in 2006. He is a recognized speaker at AI conferences and has an MBA from Stanford.

Asif Razzaq is the CEO of Marktechpost Media Inc.. As a visionary entrepreneur and engineer, Asif is committed to harnessing the potential of Artificial Intelligence for social good. His most recent endeavor is the launch of an Artificial Intelligence Media Platform, Marktechpost, which stands out for its in-depth coverage of machine learning and deep learning news that is both technically sound and easily understandable by a wide audience. The platform boasts of over 2 million monthly views, illustrating its popularity among audiences.