Reference

Nine tools,
fully described.

Most natural-language questions about papers route to one of these. Your model picks. These descriptions are what it sees in tools/list.

Terracotta ceramic open book with floating ceramic spheres above
01
search_papers

Full-text keyword search over 8.5M papers (arXiv + biomedical: PMC / bioRxiv / medRxiv, all disciplines). Typo-tolerant and ranked; returns top matches with TLDR and links. Use for topic, keyword, author, or entity queries. For 'papers like X' use similar_papers (SPECTER2).

Parameters
querystringrequired

Natural-language search query. E.g. 'transformer attention efficiency' or 'graph neural networks for molecular property prediction'.

kinteger

Number of results. Default 10, max 25.

year_mininteger

Earliest publication year, e.g. 2024.

year_maxinteger

Latest publication year.

Example prompt

"Find papers about retrieval-augmented generation published in 2024 or later."

02
get_paper

Fetch full metadata: title, authors, year, abstract, TLDR, key findings, citation count, GitHub repos, HuggingFace models and datasets, videos, peer reviews, and links.

Parameters
idstringrequired

arXiv ID like '2401.12345' or '1706.03762v5'. Also accepts OpenAlex Work IDs like 'W4390723197'.

Example prompt

"Tell me about the paper at arXiv:1706.03762."

03
similar_papers

Find papers semantically similar to a given paper using SPECTER2 (768D, trained on scientific-citation triplets).

Parameters
idstringrequired

arXiv ID or OpenAlex Work ID of the focal paper.

kinteger

Number of neighbors. Default 10, max 25.

Example prompt

"What's similar to the BERT paper?"

04
citations

Get papers that cite the given paper. Coverage for 2024+ is sparse, older years are well-indexed via OpenAlex.

Parameters
idstringrequired

arXiv ID or OpenAlex Work ID of the focal paper.

kinteger

Max citing papers. Default 25, max 100.

Example prompt

"Who cites the Attention Is All You Need paper?"

05
references

Get the papers that this paper cites (its bibliography).

Parameters
idstringrequired

arXiv ID or OpenAlex Work ID of the focal paper.

kinteger

Max references. Default 25, max 100.

Example prompt

"What does the Transformer paper cite?"

06
get_full_text

Fetch a paper's full body text as Markdown: methods, results, protocols, inline tables. Use for deep questions the abstract can't answer.

Parameters
idstringrequired

arXiv ID (2401.12345), PMC ID (PMC5339222), or bioRxiv/medRxiv DOI (10.1101/…).

Example prompt

"Summarize the methods section of PMC5946470."

07
get_figures

Get a paper's real figure images: URLs, labels, and captions. Most biomedical papers have figures; arXiv coverage varies.

Parameters
idstringrequired

arXiv ID, PMC ID, or bioRxiv/medRxiv DOI.

Example prompt

"Show me the figures from PMC5946470."

08
get_entities

Get the biomedical entities linked to a paper: genes, proteins, chemicals, diseases, species, mutations, cell lines, and clinical-trial (NCT) IDs.

Parameters
idstringrequired

arXiv ID, PMC ID, or bioRxiv/medRxiv DOI.

Example prompt

"What genes and diseases are linked to PMC5946470?"

09
get_structure

Resolve a gene/protein name or UniProt accession to its 3D structure. Returns the UniProt accession + AlphaFold model URL (and PDB when available). Great for a gene named in a paper's entities.

Parameters
namestringrequired

Gene/protein symbol (TP53, CD44) or UniProt accession (P04637).

Example prompt

"Get the 3D structure of TP53."

Navigating the API

Authentication

No API key. Add the server and it works anonymously: 50 queries / day, rate-limited by IP. When you cross that, the server replies 401 and points your client at our sign-in (OAuth 2.1 + PKCE, per the MCP spec):

HTTP/1.1 401 Unauthorized
WWW-Authenticate: Bearer resource_metadata=
  "https://mcp.tomesphere.com/.well-known/oauth-protected-resource"

Claude, Cursor, VS Code and other spec-compliant clients handle this for you: a browser opens, you sign in once, and your limit jumps to 500 / day. Nothing to paste, no card. See /mcp/install for per-client setup.

Rate limits

Three headers on every response.

X-RateLimit-Limit:      500
X-RateLimit-Remaining:  487
X-RateLimit-Reset:      2026-05-26T00:00:00Z

Over the limit returns HTTP 429 with Retry-After counting down to UTC midnight reset.

Caveats

Things worth knowing.

  • TLDRs are LLM-generated. Generated by an LLM across all 8.5M papers. Strong on average, occasionally wrong on niche papers. Flag this in user-facing claims.
  • 2024+ citation coverage is sparse. OpenAlex hasn't fully indexed recent papers. We backfill weekly. Pre-2024 is solid.
  • Some papers only have OpenAlex IDs. The Tomesphere arXiv link returns a "not indexed" page for those. The OpenAlex link still works.
Install in your client→

No key needed. Anonymous is 50 / day; sign in from your client for 500 / day.