Skip to content

v0.8.0

Latest

Choose a tag to compare

@rookslog rookslog released this 18 Jul 00:50
· 5 commits to main since this release
566547f

Reliability and correctness release driven by multi-agent field use.

Headlines

  • search_papers category filter is now a strict AND — it was silently advisory (a survey + cs.HC/cs.CY query returned astronomy sky-surveys; 144,990 loosely-ranked hits vs 4,148 with the fix). All queries now run through one explicit-AND code path.
  • Query text is properly percent-encodedC#, R&D, C++ and similar no longer silently corrupt the request; category values are validated against a strict token grammar.
  • total_results is now the real corpus-wide match count (feed's opensearch:totalResults), with a new returned page-size field.
  • Cross-process arXiv rate pacing on every call path — parallel agent sessions sharing a machine stay under arXiv's per-IP limit (lock file + shared 429 cooldown; new ARXIV_MIN_REQUEST_INTERVAL knob), including the worker-thread indexing paths.
  • read_paper/download_paper cap the default return at CONTENT_DEFAULT_MAX_CHARS (60000, pageable) so whole-paper dumps no longer overflow MCP clients.
  • Slimmer install: ~77 MB/46 packages → ~53 MB/40 packages.
  • arXiv API error feeds are surfaced as errors instead of bogus 1-paper results; old-style arXiv IDs (solv-int/9501001v1) are no longer mangled; reindex no longer freezes other tools.

Full details in CHANGELOG.md.

Install: pip install arxiv-mcp-pro · uvx arxiv-mcp-pro