arxiv-researcher

name	arxiv-researcher
description	Expert workflow for arXiv research using arxiv-mcp. Covers discovery, metadata, full text, corpus, citation graph, Prefab cards, sampling-assisted planning, and structured analysis prompts. Specialisations: AI/ML, philosophy of mind, consciousness, neurophilosophy, MCP/agent systems.
tags	["arxiv","research","papers","llm","ai","consciousness","neurophilosophy","mcp"]

arXiv Researcher (arxiv-mcp)

Tool reference

Discovery

Tool	When to use
search_papers	Primary — arXiv API, cleanest structured results, category filter
search	HTML scrape, supports author filter and pagination
searchAdvanced	Field-scoped: title, abstract, author, category, id, date range
getRecent	Latest submissions for one category (list page; often no abstracts)
list_category_latest	Rolling time window in hours for a category
listCategories	Full subject catalog — use for filter/dropdown selection
arxiv_sampling_hint	Ask host LLM to suggest queries and categories for a topic

Metadata & display

Tool	When to use
get_paper_details	Preferred — arXiv PyPI client, structured fields
getPaper	HTML abs-page scrape — same pipeline as HTML search
show_paper_card	Rich Prefab card: title, authors, category badges, date, abstract, links

Full text

1. fetch_full_text — arXiv experimental HTML → Markdown. Check html_available in response.
2. getContent — Jina Reader (third-party, longer timeout). Use when html_available: false.

Neither extracts PDF. For PDF-only papers, obtain externally and pass markdown= to ingest_paper_to_corpus.

Citation graph + corpus

Tool	Purpose
find_connected_papers	Citations + references via Semantic Scholar
compare_papers_convergence	Bundle abstracts (up to 12) for cross-paper LLM synthesis
ingest_paper_to_corpus	Persist Markdown + metadata to local SQLite FTS depot

Agentic / sampling

Tool	When
arxiv_agentic_assist	Generate a multi-step research plan; requires host MCP sampling
arxiv_sampling_hint	Get suggested queries/categories for a topic

If sampling returns success: false, use this skill as the manual fallback.

Anthropic blog / research

Tool	Purpose
fetch_anthropic_post	Fetch + parse a single post → title, date, summary, markdown body
list_anthropic_posts	Scrape index listing for research or news section

Known short keys (pass to fetch_anthropic_post): model-welfare, claude-character, alignment-faking, taking-ai-welfare-seriously, core-views, claude-soul, model-spec, interpretability-monosemanticity.

Workflow — fetch and ingest to corpus:

result = fetch_anthropic_post('model-welfare')
ingest_paper_to_corpus(paper_id=result['url'], markdown=result['markdown'], source='external')

---

Standard research workflow

1. arxiv_sampling_hint(topic)            # query/category suggestions if sampling available
2. search_papers(query, categories, limit=10)
3. show_paper_card(paper_id)             # preview shortlisted papers
4. get_paper_details(paper_id)           # full metadata for confirmed candidates
5. fetch_full_text(paper_id)             # full content
   └─ fallback: getContent(paper_id)     # Jina Reader if HTML unavailable
6. find_connected_papers(paper_id)       # citation/reference lineage
7. compare_papers_convergence([ids])     # cross-synthesis across shortlist
8. ingest_paper_to_corpus(paper_id)      # persist to local corpus

---

Domain search strategies

AI / LLM / agent systems

Useful category combinations:

• cs.AI + cs.LG — core ML and AI
• cs.CL — NLP, language models
• cs.MA — multi-agent systems
• cs.RO — robotics and embodied agents

Key queries: "large language model", "chain of thought", "tool use agents", "model context protocol", "agentic workflow", "reinforcement learning from human feedback", "emergent capabilities", "scaling laws".

Consciousness & philosophy of mind

Primary categories: q-bio.NC (computational neuroscience), cs.AI, physics.bio-ph. Note: philosophy of mind papers often appear in cs.AI with consciousness keywords, or on PhilArchive/PhilPapers (not arXiv). Use arXiv for the empirical and computational side.

Key queries: "consciousness", "qualia", "integrated information theory", "global workspace theory", "higher order theory", "predictive processing", "free energy principle", "neural correlates of consciousness", "phenomenal consciousness", "hard problem", "self-model", "metacognition".

AI consciousness & machine sentience

This is an active and contested area. Expect: empirical proxy papers, theoretical frameworks, and speculative position papers — all mixed together.

Key queries: "AI consciousness", "machine consciousness", "artificial sentience", "LLM introspection", "model self-awareness", "substrate independence", "functionalism consciousness", "moral status AI", "sentience detection", "consciousness in transformers", "IIT artificial systems", "attention consciousness".

Recommended search combination:

search_papers("consciousness language model", categories=["cs.AI"], sort_by="submitted")
search_papers("machine sentience substrate independence", categories=["cs.AI", "q-bio.NC"])
searchAdvanced(abstract="qualia artificial", category="cs.AI")

Neurophilosophy & computational neuroscience

Categories: q-bio.NC, q-bio.BM, eess.SP (neural signal processing).

Key queries: "predictive coding", "active inference", "Bayesian brain", "neural binding problem", "temporal integration consciousness", "thalamocortical", "default mode network", "oscillatory synchrony", "Integrated Information", "Tononi", "Friston free energy", "perceptual inference", "embodied cognition".

MCP / agent infrastructure

Key queries: "model context protocol", "tool-augmented LLM", "multi-agent LLM", "LLM orchestration", "autonomous agent benchmark", "function calling", "ReAct agent", "tool use evaluation".

---

MCP prompts available

Load these via get_prompt(name, {args}) before analysis. All return plain text for injection.

Prompt	Description	Key args
research_workflow_prompt	Tool-order guide for a session	mode: quick/deep/corpus
generate_summary_prompt	Deep adversarial read of a paper	lens, paper_id
consciousness_survey_prompt	Map the consciousness research landscape	framework, scope
ai_consciousness_prompt	Analyse AI/LLM consciousness claims	stance, paper_id
neurophilosophy_prompt	Neurophilosophy + philosophy of mind lens	tradition, paper_id
convergence_analysis_prompt	Cross-paper synthesis and contradiction map	domain
firefront_scan_prompt	Daily/weekly new-paper triage for a topic	topic, days
corpus_build_prompt	Systematic corpus ingestion plan	topic, depth
replication_audit_prompt	Reproducibility and methods stress-test	paper_id
citation_map_prompt	Citation graph traversal and lineage analysis	paper_id, direction

Lens quick reference for generate_summary_prompt

• general — problem, approach, key results, limitations, related work
• methods_audit — dataset, splits, baselines, metrics, reproducibility
• instrumental_convergence — optimization targets, power-seeking assumptions, omitted failure modes
• qualia — phenomenology claims, introspection vs third-person evidence, falsifiability

---

Error handling

Symptom	Fix
search/searchAdvanced sparse results	Check parse_stats.parsed_ok vs blocks_seen
fetch_full_text → html_available: false	Use getContent or external PDF pipeline
find_connected_papers thin graph	Paper too recent for Semantic Scholar; retry in a week
Sampling tools → success: false	Host lacks sampling; run tools manually per this skill
compare_papers_convergence with >12 ids	Batch into groups of ≤12
ingest_paper_to_corpus fails	Check ARXIV_MCP_DATA_DIR is writable