paper-research

Paper Research Pipeline

Overview

Orchestrates the full academic paper workflow: search/download → Zotero import → Obsidian output. Guides the user through each stage and calls the appropriate sub-skills. Each sub-skill can also be used standalone.

Installed MCP Tools

These are already configured — do not prompt user to install them:

MCP	Purpose
openags/paper-search-mcp	Search arXiv, PubMed, Semantic Scholar + download PDFs
mcp-zotero (Xevos117)	Import papers to Zotero by DOI or PDF
ajtruex/mcp-obsidian-streamable-http	Read/write Obsidian vault

Sub-skills

Sub-skill	Purpose	Standalone?
paper-fetch	Search + select + download papers	✓
paper-zotero	Import to Zotero library	✓
paper-note	Structured Obsidian note per paper	✓
paper-moc	Topic MOC index page	✓
paper-wiki	Karpathy cumulative wiki	✓
paper-dataview	YAML frontmatter for Dataview queries	✓

Pipeline Flow

Stage 1 — Fetch Call paper-fetch. Present candidate list. Wait for user selection.

Stage 2 — Zotero Import Call paper-zotero with:

• Selected papers from Stage 1
• Collection name = search topic (e.g. "RAG for Code Generation") — create automatically, do not ask user

After import, run the PDF fallback step inside paper-zotero: for each paper with no PDF attached, try import_pdf_to_zotero using the local path from Stage 1.

Stage 3 — Obsidian Output Ask the user: "Which Obsidian output would you like? (choose one or more)"

• A) Single notes — one structured note per paper (paper-note)
• B) Topic index — MOC page aggregating papers (paper-moc)
• C) Wiki — Karpathy-style deep wiki: chapter-by-chapter paper breakdown + detailed concept entries (paper-wiki). Reads the full PDF — not just the abstract. Each paper page 600–1200 words; each concept entry 400–800 words.
• D) Dataview metadata — YAML frontmatter for queries (paper-dataview)
• E) Skip Obsidian — Zotero only

paper-dataview runs automatically alongside A, B, or C unless user opts out.

Quick Mode

• DOI list provided → skip Stage 1, go straight to Stage 2
• PDF only, no DOI → skip Stage 1 + Stage 2 Zotero metadata step, use import_pdf_to_zotero directly, then go to Stage 3
• Summarize only → skip Stage 1 + Stage 2, run paper-note directly on given PDF

Example Interactions

"Research the latest papers on RAG for code generation" → Keyword search → present list → user selects → import to Zotero collection "RAG for Code Generation" → ask Obsidian preference

"Import these DOIs to Zotero and create wiki entries" → Skip fetch → import to Zotero → run paper-wiki

"Summarize this PDF and add it to my Obsidian" → Skip fetch + Zotero → run paper-note directly

Error Handling

• If paper-search-mcp is unavailable, fall back to asking the user for a DOI list
• If Zotero import fails for a paper, log it and continue — do not abort the batch
• If Obsidian MCP is unavailable, output the note content inline in the chat instead