Pod

Available as Markdown and JSON. Pod is also available over MCP.

Reported issues for cern-inspire-mcp-server

Pod holds 12 of 12 GitHub reports that passed its relevance review. This can include external user reports, maintainer-confirmed bugs, and concrete feature gaps. Treat them as evidence to inspect, not a count of distinct defects.

Back to cern-inspire-mcp-server.

Most discussed

feat(cern_inspire_search_authors): flag name results where no candidate carries the queried surname

Use case

INSPIRE's free-text author search matches name parts as initials. A misspelled or unknown surname in cern_inspire_search_authors therefore still returns hundreds of thousands of unrelated profiles, and none of them carries the surname. The tool's zero-hit name notice ("Try "Last, First", fewer name parts…") never fires. The only guidance is "raise limit", so a caller can take an unrelated namesake-by-initial for the person it asked about.

Repro (server 0.1.0, mcp-ts-core…

Read the thread · 2026-10-01 · closed · 2 comments

bug(cern_inspire_get_paper): truncation notice says to raise max_authors when it is already at the 500 maximum

Server version

0.1.0

mcp-ts-core version

0.13.10

Runtime

Node.js

Runtime version

Node 26.5.0

Transport

HTTP (Streamable HTTP)

Description

When max_authors is already at its maximum (500), the truncation notice in cern_inspire_get_paper still says to raise it. A retry with a higher value then fails validation (max_authors: Too big: expected number to be <=500). The notice should only point at max_authors while there is headroom. At the cap, it should…

Read the thread · 2026-10-01 · closed · 2 comments

bug(cern_inspire_search_experiments): reports "Ongoing: no" for active experiments with no completion date

Server version

0.1.0

mcp-ts-core version

0.13.10

Runtime

Node.js

Runtime version

Node 26.5.0

Transport

HTTP (Streamable HTTP)

Description

cern_inspire_search_experiments reports ongoing: false and renders **Ongoing:** no for any experiment whose INSPIRE record has no date_completed. INSPIRE leaves that field out for many active experiments, so DUNE and EUCLID are reported as not ongoing. ongoing is only known when the record carries the "9999"…

Read the thread · 2026-10-01 · closed · 2 comments

bug(cern_inspire_get_paper): a merged recid returns paper_not_found instead of the record INSPIRE redirects to

cern_inspire_get_paper looks a recid up with a recid:N search. INSPIRE drops a merged record from search results but keeps its ID as a redirect, so an old recid fails as paper_not_found even though INSPIRE still serves the paper.

Repro (2026-10-02):

Read the thread · 2026-10-02 · closed · 1 comment

bug(notices): angle-bracket query placeholders vanish when content[] renders as HTML

Server version

0.1.1

mcp-ts-core version

0.13.10

Runtime

Bun

Runtime version

1.4.2

Transport

HTTP (Streamable HTTP)

Description

Several notices and recovery hints the server writes itself spell query placeholders with angle brackets: "t <words>", "a <name>", collaborations.value:<name>, "collaboration:<name>", "a <BAI or name>". These reach content[] as written. CommonMark reads each as an inline raw-HTML open tag (<BAI or name> is tag BAI with…

Read the thread · 2026-10-02 · closed · 1 comment

bug(inputs): advertised inputSchema rejects paper and facet spellings the descriptions document

Server version

0.1.1

mcp-ts-core version

0.13.10

Runtime

Bun

Runtime version

1.4.2

Transport

HTTP (Streamable HTTP)

Description

paperInput, documentTypesInput, and subjectsInput normalize in a z.preprocess, but inputSchema advertises only the normalized form. paper's pattern rejects every prefixed, versioned, URL, and ins<recid> spelling its description lists; document_types and subjects (cern_inspire_search_literature,…

Read the thread · 2026-10-01 · closed · 1 comment

feat(cern_inspire_get_citation_summary): add citations per year from INSPIRE's citations-by-year facet

Use case

cern_inspire_get_citation_summary returns totals but no trajectory. Whether a paper is still being cited, or an author's citations per year for a tenure or grant dossier, takes one literature search per year today. INSPIRE serves the per-year series from the /api/literature/facets endpoint this tool already calls (facet_name=citations-by-year).

Proposed behavior

Request facet_name=citations-by-year in parallel with citation-summary, with the same q, and return…

Read the thread · 2026-10-01 · closed · 1 comment

feat(cern_inspire_list_reference): document affiliation queries and point institution recids at them

Use case

No documented path leads to "papers from an institution" or "an institution's citation summary." cern_inspire_get_citation_summary lists an institution as a valid query. cern_inspire_search_authors (institutionRecid) and cern_inspire_search_experiments (institutions[].recid) return institution recids, but nothing says which query takes them, and search_syntax lists no affiliation operator.

Proposed behavior

Read the thread · 2026-10-01 · closed · 1 comment

Most recent

feat(cern_inspire_export_citations): add page so citation sets over 50 entries can be exported in full

Use case

A paper's reference list or citing set often exceeds cern_inspire_export_citations' 50-entry size cap, and the tool takes no page: citedby:recid:1124337 (the references of the ATLAS Higgs observation) matched 138 papers on 2026-10-01, and only the first 50 can be exported. The workaround is paging cern_inspire_search_literature and re-exporting the recids in OR batches.

Proposed behavior

Add page (1-based, default 1; page × size ≤ 10,000, else…

Read the thread · 2026-10-01 · closed · 1 comment

bug(normalize): titles and abstracts keep publisher HTML/MathML markup and undecoded entities

Server version

0.1.1

mcp-ts-core version

0.13.10

Runtime

Bun

Runtime version

1.4.2

Transport

HTTP (Streamable HTTP)

Description

INSPIRE stores many publisher titles and abstracts as HTML, JATS, or MathML, and HEPData abstracts entity-escaped (&lt;). normalize.ts copies them verbatim: structuredContent carries the markup, content[] renders it escaped, and abstractSnippet is cut from the raw string. As of 2026-10-01, titles.title:mrow matches 5,971…

Read the thread · 2026-10-01 · closed · 1 comment

bug(normalize): hepdataUrl can open a different HEPData record than the DOI and table count it reports

Server version

0.1.1

mcp-ts-core version

0.13.10

Runtime

Bun

Runtime version

1.4.2

Transport

HTTP (Streamable HTTP)

Description

cern_inspire_get_paper and cern_inspire_search_hepdata build hepdataUrl as https://www.hepdata.net/record/ins<paper recid>. hepdata.net resolves ins<N> by the INSPIRE ID stored on the HEPData record, which stays on the old recid when INSPIRE merges that paper into another and re-links the data. The link can then open a…

Read the thread · 2026-10-01 · closed · 1 comment

feat: read HEPData table values once hepdata.net admits the server

Add the two tools that read HEPData table values from hepdata.net: cern_inspire_list_hepdata_tables (a record's tables, with names, descriptions, keywords and download formats) and cern_inspire_get_hepdata_table (one table's headers, qualifiers, values and uncertainties). Their specification is in docs/design.md under Deferred: HEPData-direct tools.

Blocked on access. hepdata.net serves a Cloudflare managed challenge (403, cf-mitigated: challenge) to the server's own…

Read the thread · 2026-10-01 · open · 0 comments

The remaining reports are on the project's issue tracker.