Skip to content

Fix paper-search skill to use connected MCP servers - #2

Open
conormccallion1-sys wants to merge 1 commit into
mainfrom
claude/determined-wright-2vz8ad
Open

Fix paper-search skill to use connected MCP servers#2
conormccallion1-sys wants to merge 1 commit into
mainfrom
claude/determined-wright-2vz8ad

Conversation

@conormccallion1-sys

Copy link
Copy Markdown
Owner

Problem

The paper-search skill was broken. SKILL.md instructed Claude to run a CLI via uv run --directory <REPO_PATH> paper-search …, but:

  • No such package exists in the repo — it only contains README.md and SKILL.md, no pyproject.toml or source.
  • It couldn't run here even if built — the sandboxed environment blocks direct egress to academic APIs (arxiv.org, crossref.org, ncbi.nlm.nih.gov, openalex.org, etc. all return "Host not in allowlist"), and a uv run subprocess can't reach the session's MCP servers either.

Meanwhile, the connected research MCP servers already cover the same sources and work (verified live: Consensus returned 20 papers across Semantic Scholar/PubMed/Scopus/arXiv; PubMed matched 11k+ articles).

Fix

Rewrite the skill to drive the working MCP tools instead of the phantom CLI:

  • SKILL.md — routing table mapping each job to a real tool:
    • Search → Consensus, Scholar_Gateway, PubMed, bioRxiv, Exa
    • Read full text → PubMed get_full_text_article / get_article_metadata, bioRxiv get_preprint, Exa web_fetch
    • Related / citations → PubMed find_related_articles / lookup_article_by_citation / convert_article_ids
    • Plus a workflow and a note explaining why there's no CLI.
  • README.md — now describes the repo and usage instead of placeholder text.

No egress required: the MCP servers run outside the sandbox's network boundary.

🤖 Generated with Claude Code

https://claude.ai/code/session_01EXmEUy9tbicVu62VatGrE3


Generated by Claude Code

The skill pointed to a `paper-search` CLI run via `uv run`, but no such
package exists in the repo and sandboxed sessions block egress to academic
APIs, so a local HTTP client could never work. Rewrite SKILL.md to drive the
already-connected research MCP servers (Consensus, PubMed, bioRxiv/medRxiv,
Scholar Gateway, Exa), which cover the same sources without needing egress.
Map each job (search / read full text / related) to the right MCP tool and
document the workflow. Update README to match.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EXmEUy9tbicVu62VatGrE3
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants