Scholar MCP 1.10 replaces its two separate background-task systems with one:
every tool whose work can run long now promotes to a background job on the
same schedule and is polled with a single new tool, get_job_result. The
patent tools gain a real retry against EPO's throttling instead of one that
never actually asked EPO again, two gaps in the advertised tool metadata are
closed, and fetch_and_convert stops re-running its most expensive step when
a cached result already exists.
Features
- wire the pvl-core Jobs framework and move the PDF tools onto it (#320)
- move the patent tools onto Jobs, with real EPO backoff (#321)
- move the paper, book and citation tools onto Jobs (#324)
- move the citation-graph and utility tools onto Jobs (#325)
- delete the bespoke task queue and consolidate on one contract (#326)
Bug Fixes
-
shorten domain_description to fit the MCP registry's 100-char cap (#282)
-
gate the tool-registration metadata and close two surface defects (#328)
-
reuse the cached Markdown in fetch_and_convert (#329)
-
Release notes: https://pvliesdonk.github.io/scholar-mcp/1.10/releases/1.10/
-
Documentation: https://pvliesdonk.github.io/scholar-mcp/1.10/
-
All changes since v1.9.1: v1.9.1...v1.10.0
-
Commit-level detail: CHANGELOG.md