That is the whole design: long pages and PDFs get compressed by a separate OpenAI-compatible endpoint before they land in your assistant's context, and you choose the ratio — 0.2 means a fifth of the original. It also means a second API key of your own, and file and page content is capped at roughly 120k characters going in.
A summarization server: it fetches pages and reads files, then runs them through an OpenAI-compatible model you configure and returns the condensed version.
- Fetch a web page, optionally saving it to a txt file — scrape_webpage
- Summarize any text to a target ratio, 0.2 by default — summarize_content
- Fetch and summarize a page in one call — summarize_webpage
- Summarize a txt file or a PDF from a path — read_and_summarize_text_file, read_and_summarize_pdf_file
- Pull the parts of a document that answer a question, returned within 2k characters — topic_based_summary
- Send a prompt directly to the configured model — call_model
An API key in OPENAI_API_KEY — MiniMax by default, with OPENAI_BASE_URL and OPENAI_MODEL pointing anywhere OpenAI-compatible. Install with uvx fullscope-mcp-server or pip install fullscope-mcp-server; version 1.0.1.
One command plus a key — uvx fullscope-mcp-server, then supply credentials