MCP server that enables AI assistants to read arXiv papers and save notes to Notion
Server has 6 tools with complete schemas and descriptions. Naming follows verb_noun convention (read_, get_, search_, save_, update_). Descriptions are detailed (100-300+ chars) and explain WHEN to use each tool. However, several critical gaps: (1) save_paper_to_notion and update_paper_in_notion lack output schema documentation, LLMs cannot plan downstream steps or extract returned IDs; (2) parameter descriptions are verbose but lack formal constraints (enums, ranges, patterns), e.g., 'categories' accepts arbitrary strings instead of a constrained list of valid arXiv categories; (3) no error handling guidance, tools do not document what happens on failure (invalid paper_id, Notion auth failure, missing figure); (4) no idempotency hints for write operations; (5) figure_path parameter in save_paper_to_notion requires manual LaTeX parsing, creating a fragile UX. Read-only tools (read_arxiv_paper, get_daily_digest, search_by_tags, get_existing_tags) are well-structured. Write tools lack recovery guidance.
Get the latest arXiv papers from specified categories. Returns a list of recently submitted papers with their titles and abstracts. Useful for staying up to date with new research.
Get a list of all existing tags used in the Notion Papers database. Use this before saving/updating a paper to choose appropriate tags from the existing list. Returns a list of tag names.
Read an arXiv paper by its ID. Downloads the LaTeX source and returns: 1) Full paper content (title, authors, abstract, complete text), 2) Absolute paths to all figures/images in the paper, 3) Paths to source files. Use the returned image paths to view paper figures if needed.
Save a NEW arXiv paper to the Papers database (Insert Mode). Automatically fetches metadata. You MUST provide: 1. paper_id: The arXiv ID. 2. chinese_summary: A one-sentence Chinese summary to prepend to the abstract. 3. tags: A list of tags (prefer using existing ones from get_existing_tags). 4. notes: DETAILED notes IN CHINESE (中文), using Markdown format. Must include: - 创新点/Insight (论文的核心贡献) - 研究背景 (问题是什么,为什么重要) - 现有方法的不足 - 本文方法 (方法细节) - 实验结果 - 结论与启发 5. figure_path: ALWAYS include Figure-1. Read the paper's LaTeX source to find the \includegraphics command for Figure 1, then provide the absolute path to that file.
Write tools (save_paper_to_notion, update_paper_in_notion) lack output schema documentation. LLMs cannot extract returned page_id or status to chain downstream operations.
Parameter 'categories' in get_daily_digest and search_by_tags accepts free-form strings instead of an enum of valid arXiv categories (cs.AI, cs.LG, cs.CL, etc.). LLMs will hallucinate invalid category names.
No error handling guidance. Tools do not document what happens on invalid paper_id, Notion auth failure, missing figure file, or API timeouts. LLMs have no recovery path.
figure_path parameter requires LLM to manually parse LaTeX source and locate \includegraphics commands. This is fragile and error-prone. Should be automated or offer a helper tool.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | <=2025-11-25 | v2 |
Search arXiv for papers matching specific research topics or tags. Returns relevant papers with titles and abstracts. Useful for finding papers related to specific research interests.
Update an EXISTING paper in Notion (Edit Mode). Use this to fill in missing information for a paper that is already in the database. Can update the abstract (prepend summary), tags, and append notes.
No idempotency or confirmation hints for destructive write operations. LLMs may retry save_paper_to_notion on ambiguous failures, creating duplicate Notion entries.