paper-search-agent

English | 中文

A local-first scholarly paper discovery, access planning, and full-text retrieval agent.

• Multi-source search with automatic de-duplication via search_papers.
• Single-source deep search via search_single_source.
• Enabled discovery sources in current codebase:
• OpenAlex
• Crossref
• Scopus
• Springer Meta
• arXiv
• PubMed
• Europe PMC
• Unpaywall

• DOI/paper resolution and route planning via resolve_and_plan.
• Local cache checks via check_local_cache.
• Route preference order: local cache -> Zotero -> OA -> publisher APIs -> browser-assisted -> manual import.

• Unified retrieval execution via fetch_fulltext.
• Human-in-the-loop browser retrieval via browser_retrieve.
• Local file ingestion via import_local_file.

• Artifact parsing (xml, html, pdf, text) via parse_paper.
• Section-level extraction via get_paper_sections.
• Soft token-budget advisory (no hard truncation).

• Corpus management via manage_corpus (add, list, remove, deduplicate).
• Export in json, csv, bibtex via export_records.

• Zotero tools (when enabled): zotero_lookup, zotero_save, zotero_list_collections.

The practical pipeline is:

1. Discover papers (search_papers / search_single_source) -> produce CandidatePaper[].
2. Plan access (resolve_and_plan) -> produce AccessPlan.
3. Retrieve artifact (fetch_fulltext / browser_retrieve / import_local_file) -> store local PDF/XML/HTML/text.
4. Parse content (parse_paper) -> produce NormalizedPaperRecord.
5. Read targeted sections (get_paper_sections) and optionally save/export (manage_corpus, export_records).

This separation is intentional: discovery does not imply full-text entitlement.