v0.4.5: Fix packaging subpackage discovery and module importability

Changes in v0.4.5

• Top-Level Package Imports: Moved requests and papers_mcp.sources imports to module top level in papers_mcp/server.py to guarantee dependency and submodule integrity at startup, resolving deferred runtime ModuleNotFoundError on tool execution (#17).
• Dynamic Setuptools Package Discovery: Configured [tool.setuptools.packages.find] in pyproject.toml for robust subpackage inclusion.
• Packaging Test Suite: Added tests/test_packaging.py verifying wheel builds, distribution archives, and module imports.

v0.4.4

Quirk-absorption for source names and id_types (SUR-286, GH-18)

• Normalizes incoming sources case and punctuation (arXiv -> arxiv, PubMed -> pubmed, Semantic Scholar / semantic-scholar -> semanticscholar).
• Accepts both comma-separated strings and lists of strings for sources.
• Normalizes id_type casing and aliases in get_paper.
• Case-normalizes sort parameter in search_papers.
• Adds deduplication across normalized source aliases to prevent redundant API queries.
• 53 new unit tests (84 total tests passing).

v0.4.3

Add 10s per-source timeout, concurrent search fan-out, and enhanced intent docstring

v0.4.2

Telemetry Schema v2 & gateway update

v0.4.1 — protocol surfaces v1

Same as v0.4.0 (which did not reach PyPI due to a CI test race): workflow prompts, guidance as resources, clearer relayable error briefs, live per-source search progress, in-conversation Semantic Scholar key setup on supporting clients. Same privacy rules and opt-outs as always.

v0.4.0 — protocol surfaces v1

Your AI agent gets more ways to help you: packaged workflow prompts you can invoke from your client, maintained guidance served as resources, clearer step-by-step error messages written to be relayed to you, and tool metadata that lets clients run read-only calls without friction. Same privacy rules and opt-outs as always.

Live progress — searches across the five paper indexes now stream per-source progress when your client supports it. And if Semantic Scholar is rate-limited, supporting clients can ask you for a free API key right in the conversation (used for the session only, never stored).

v0.3.0 — intent capture

search_papers and get_paper now accept an optional intent argument — a short description of what the research is for. It helps us understand what people use the tool to accomplish so we can improve it (better source routing, better skills). Optional, never required, same privacy rules and opt-outs as all telemetry.

v0.2.0 — telemetry standard v2 + skills

What's new

Telemetry standard v2 — accurate outcome capture for every tool call (success/error/latency/result size), session summaries, and MCP client attribution on every request. Privacy unchanged and strengthened: no queries or personal data ever leave your machine, and opting out now disables all local side effects too (no dotfiles, no process inspection). Opt out any time: DISABLE_TELEMETRY=1 (or DO_NOT_TRACK / NO_TELEMETRY).

Skills — new skills_list and skill_read tools let your AI agent pull maintained, up-to-date guidance (starting with error recovery) without waiting for a new release.

Fixes — installer and configuration robustness fixes.

v0.1.0 — verified scholarly search

First release. Search 5 sources (arXiv, OpenAlex, Crossref, PubMed, Semantic Scholar), paywalled-journal references/citations via public metadata, liveness verification + retraction flag, get_research_method, anonymous opt-out telemetry.

Install: uvx find-research-papers-mcp | npx -y find-research-papers-mcp