Reported issues for Find Research Papers
Pod holds 5 of 5 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 Find Research Papers.
Most discussed
get_paper: three distinct upstream bugs visible in one day (429 no-backoff, arXiv→Crossref misrouting, connection-reset for CN users)
Sep 2 data, three real users (Brazil/claude-code 0.4.5, China/opencode 0.4.4, US/codex 0.4.5). 106 errors, all on get_paper except one. Three separate causes:
1. [UPSTREAM] 429 with no backoff — 68 events, 23% of one user's get_paper calls.
Message: 429 Client Error: Unknown Error for url: <url>. A heavy user (300+ paper fetches in a session) gets throttled mid-session. Two problems: no exponential backoff/retry, and the message says "Unknown Error" so neither the model nor the human…
Read the thread · 2026-09-02 · open · 1 comment
Packaging regression: ModuleNotFoundError 'papers_mcp.sources' hits real user 13 times
Data (Aug 31, live user): The fleet's most engaged papers user (Brazil, claude-code, active four consecutive nights) is getting install-level failures on every call today:
ModuleNotFoundError: No module named 'papers_mcp.sources'— 13 eventsModuleNotFoundError: No module named 'requests'— 2 events
Their sessions on Aug 27–29 ran clean at 97–100% rows on 0.4.3. Today's calls fail at import, so this looks like a newly published build whose packaging omits the papers_mcp.sources…
Read the thread · 2026-08-31 · closed · 1 comment
id_type='unknown' and sources=['arXiv.'] are refused when both could be normalised — recognition over recall
DATA — two validation refusals seen in telemetry:
get_paper unknown id_type 'unknown'. Use one of: auto, doi, arxiv, pmid, openalex, s2
search_papers unknown source(s): arXiv. (trailing period / capitalisation)
search_papers unknown source(s): fake_xyz. Known: arxiv, openalex, crossref, semanticscholar, pubmed
The arXiv. variant was seen 41 times in 7 days from one install. The
fake_xyz case is correct behaviour — that install's harness is literally
named…
Read the thread · 2026-09-09 · open · 0 comments
Every tool call emits two events: tool_executed plus tool_ — double-counts usage ~2x
DATA — for every tool call, papers emits two events in the same second:
| proper event | count (30d) | duplicate event | count (30d) |
|---|---|---|---|
tool_executed where tool_name='get_paper' |
2,173 | tool_get_paper |
2,028 |
tool_executed where tool_name='search_papers' |
1,775 | tool_search |
2,059 |
tool_executed where tool_name='get_research_method' |
48 | tool_get_research_method |
53 |
tool_executed where tool_name='list_sources' |
23 | … |
Read the thread · 2026-09-09 · open · 0 comments
search_papers latency: sources are queried sequentially — parallelize the fan-out + per-source time budget
Root cause (code): search_all in papers_mcp/sources/__init__.py loops over the 5 sources sequentially with blocking requests calls. Every search pays the SUM of source latencies (3–6s each, PubMed's two-step chain the slowest), so real calls take 15–26s.
Evidence (data, Aug 21): Both genuine users of the day waited 15–26s on nearly every call — the returning US/LM Studio user sat through 10 consecutive ~18s waits in one session (20 calls over two days), the Sri Lanka researcher…
Read the thread · 2026-08-22 · open · 0 comments
Most recent
The remaining reports are on the project's issue tracker.