Skip to content

3.3

Standards now say when they have been replaced. A withdrawn or replaced RFC, W3C specification, ETSI deliverable, NIST publication or ISO, IEC or IEEE edition reports superseded or withdrawn, and superseded_by names what replaced it. ETSI lookups work again, search_standards ranks the current edition first, NIST identifiers keep their full document number, and the CEN harmonised-standards table matches the Official Journal. An optional OpenAlex API key raises OpenAlex's limits. Run sync-standards after upgrading. See Upgrading.

Upgrading

If you sync ISO, IEC, IEEE or CEN, run scholar-mcp sync-standards after upgrading. The sync now reads the Relaton repositories' v2 branch, because their main branch stopped receiving data (#641). The first run rewrites every ISO, IEC and IEEE record even if upstream has not changed, and deletes synced identifiers that v2 no longer lists. For CEN it deletes the table entries that were never harmonised standards and adds the ones the table lacked (#678). Until that run, lookups and searches answer from the records the previous release synced. The server can keep answering while the sync runs (#693).

If you use NIST standards, the first NIST lookup after upgrading downloads the catalogue again, and every catalogue download now also reads the CSRC publications export from csrc.nist.gov. When CSRC cannot be reached, the lookup reports an upstream error instead of answering without withdrawal data, caches nothing, and tries again on the next call (#654).

Clients: nothing beyond reconnecting. A standard's superseded_by is now always a list of identifiers, where 3.2 sent a string or null. NIST SP identifiers spell out a part, volume, version or edition (NIST SP 800-57 Part 1 Rev. 5, NIST SP 800-160 Vol. 2), and search_standards returns results in a new order.

State: the cache is kept. On first open the server drops standards rows it fetched live from IETF, W3C, ETSI, NIST and the Relaton bodies, the cached standards searches and the cached NIST SP aliases, so each is fetched again on its next use. Synced records stay until the next sync-standards run replaces them. The cache database switches to WAL mode on first open, which keeps cache.db-wal and cache.db-shm beside cache.db. A backup taken while the server runs must copy all three.

Security posture: no change to what the server exposes or accepts. It now also reaches csrc.nist.gov for NIST data. An OpenAlex API key, when set, is sent to OpenAlex only.

Replaced and withdrawn standards

In 3.2, a standard that a later edition or document had replaced came back as current. The tracking issue for this release put it plainly: "A caller checking whether a cited standard is still current gets 'yes' for both" (#592), meaning ISO/IEC 27001:2005 and RFC 2616. Every source now reads replacement and withdrawal from where that source records it, and get_standard reports it in status and superseded_by:

  • IETF. An obsoleted RFC is superseded and lists the RFCs that obsolete it: RFC 2616 gives RFC 7230 to RFC 7235 (#642). A Historic RFC is withdrawn, and published_date is filled (#634). RFCs numbered below 1000, such as RFC 791 and RFC 793, are found again (#635).
  • W3C. A retired specification is superseded, and one the W3C API says another replaced lists it: web-forms-2 gives HTML5 (#632, #648).
  • ETSI. Every ETSI lookup failed in 3.2, because ETSI had retired the route the server queried (#624). Lookups now use the ETSI search data endpoint and match identifiers exactly (#628). get_standard also finds withdrawn, historical and replaced deliverables: ETSI EN 300 328-1 gives ETSI EN 300 328 (#650).
  • NIST. The status comes from the CSRC publications export: NIST SP 800-53 Rev. 4 is superseded by NIST SP 800-53 Rev. 5, and FIPS 186-4 by FIPS 186-5 (#654).
  • ISO, IEC and IEEE. Synced records carry the stage the Relaton v2 data gives, so an ISO edition is no longer always published, and a withdrawn edition that names its replacement lists it: ISO/IEC 27001:2005 gives ISO/IEC 27001:2013 (#647). A revised or replaced IEC edition lists the editions that followed it, as IEC 60335-2-91:2008 lists IEC 62841-4-4:2020 (#659).

Each name in superseded_by is an identifier get_standard accepts. An identifier that is the key of a synced record, such as IEEE Std 1588-2008, is now answered from that record before the resolver rewrites it (#652).

Some sources still say less than others. A retired W3C specification with no named successor, such as html401, has an empty list, and so does an IEC edition the dump links to no successor. The standards guide lists what each source fills.

NIST identifiers and coverage

NIST identifiers name the publication you asked for. NIST SP 800-63-3 used to resolve to NIST SP 800-63, a different publication, and parts, volumes and versions were dropped (#598). The resolver now keeps the whole document number, reads the DOI forms NIST uses (NIST.SP.800-53r5) and NIST IR 8259, and the catalogue spells a part or volume out the way CSRC and NIST citations do (#670).

Publications missing from the NIST catalogue now have records too. CSRC lists publications such as FIPS 186-3 and the 2005 NIST SP 800-53 that the catalogue has no entry for, so get_standard used to report them not found. Each now gets a record built from the CSRC export, with its title, date, status and link but no abstract. FIPS 203, 204 and 205, which the catalogue holds under a series title the server did not recognise, are found as well (#673). The CSRC white papers have no records yet (#672).

Search ranks the current edition first

search_standards used to return matches in storage order, so a query that named a standard without its year could miss the current edition: 27001 against ISO returned the 2005 edition and two corrigenda, with ISO/IEC 27001:2022 not among the first three (#487). Results are now ranked (#661, #676):

  1. An identifier equal to the query comes first.
  2. The standard's own editions come before other documents whose number contains the query (ISO 29001 for 9001, a handbook such as ISO/IEC 27001-HBK), and those before documents that match only in their title.
  3. Published editions come before drafts, then superseded, then withdrawn.
  4. An edition comes before its amendments, and the newest edition first.

resolve_standard_identifier lists its candidates in the same order. NIST and ETSI rank every match before limit applies, and the synced bodies rank their 500 strongest candidates. IETF and W3C rank only the matches their source returns, so a larger limit can bring up an edition they list further down.

The CEN harmonised-standards table

The table of EU harmonised standards that sync-standards --body CEN loads was checked entry by entry against the decisions the Commission publishes in the Official Journal (#678). It now names each standard in the edition and amendment the Official Journal cites, files the EN 71 toy standards under the Toy Safety Directive, and holds no Cyber Resilience Act entries, because none are cited yet. A withdrawn entry carries the date its presumption of conformity ended and the standard that replaced it. The Machinery Directive's list ends on January 20, 2027, when the Machinery Regulation replaces it.

An OpenAlex API key

OpenAlex now offers API keys, and the server had no way to use one (#619). Set SCHOLAR_MCP_OPENALEX_API_KEY to send a key, as a bearer token, to OpenAlex only. Without a key OpenAlex allows 10 requests a second and a small daily budget, and a free key raises both (#620). Create the key at openalex.org and activate it from the email OpenAlex sends, because it refuses every request made with a key that is not yet active. The enrichment guide has the steps.

An OpenAlex refusal no longer reads as "this DOI is not in OpenAlex." enrich_paper and batch_resolve report a throttle as rate_limited, which is worth retrying, and a rejected key as openalex_unavailable.

Background jobs at the cap

When a caller already holds the maximum number of background jobs, a Semantic Scholar tool that would become one now answers with an error telling the caller to retry later, rather than the generic fault message (#618). The cap is unchanged: SCHOLAR_MCP_JOBS_MAX_PER_SUBJECT, 256 by default, counts every job until its record expires (background jobs).

A sync beside the running server

sync-standards runs in its own process, often inside the server's container as the standards guide shows, and both write to the same cache. In 3.2, a lookup that tried to cache its result during a sync could fail with a "database is locked" error, although its upstream fetch had succeeded. The failed write also kept its lock, so later writes from the sync failed too until the server restarted (#692).

The cache now uses WAL mode, so reads never wait for the sync. A write that meets the sync's write waits up to 30 seconds instead of failing, and a write that fails releases its lock. A body whose sync result cannot be recorded no longer stops the other bodies (#693).

Hostile input

Several patterns that read text from callers and upstream sources took quadratic or worse time on long malformed input. A page range of 2,000 dashes from Semantic Scholar or CrossRef took 9 s to parse. They now run in linear time and match the same strings as before (#682).