Research Lookup

Research Lookup is the evidence-gathering skill in K-Dense's Claude Scientific Writer (MIT). Older listings described it as a Perplexity Sonar lookup. Version 1.4 changed that: it now uses Parallel Search and Extract by default, and Perplexity (via OpenRouter) only when you ask for it or allow it as a failure fallback. The default academic workflow targets 60 verified, unique references and writes a manuscript-ready research packet instead of a loose list of links.

Key Features

• Multi-pass academic search: bounded passes for recent primary studies, systematic reviews and meta-analyses, seminal papers, methods and benchmarks, and contradictory or null results, plus an unrestricted companion pass to reduce blind spots.
• Verification with Extract: candidates are deduplicated and ranked, then Parallel Extract pulls authors, venue, DOI/PMID, design, sample size, findings, limitations, and retraction or preprint status. Search-only records are not counted as verified.
• Research packet: --packet-dir writes packet.md, references.bib, an evidence matrix, a claim-to-source map, synthesis notes, section briefs for Introduction and Discussion, a coverage report, and a search ledger with exact queries and timestamps.
• Other backends: --force-backend research for explicit deep research, --force-backend chat for Parallel Chat, --force-backend perplexity for Perplexity, and --no-academic for a quick bounded lookup.

Use Cases

• Gathering background, methods precedent, and conflicting evidence for a paper's Introduction and Discussion.
• Building a claim-to-source map before drafting with Scientific Writing.
• Checking whether a statistic or recent finding is backed by a primary source.

Pricing and Access

The skill is free (MIT), but queries are billed by Parallel (log in with parallel-cli login or set PARALLEL_API_KEY). Perplexity use needs OPENROUTER_API_KEY. Query text is sent to those services, so the skill says not to use it for unpublished or private material.

Getting Started

1. Install the plugin in Claude Code: /plugin marketplace add https://github.com/K-Dense-AI/claude-scientific-writer, then /plugin install claude-scientific-writer.
2. uv tool install "parallel-web-tools[cli]==0.7.1", then parallel-cli login.
3. Run python skills/research-lookup/scripts/research_lookup.py "your research question" --academic --target-references 60 --packet-dir sources/manuscript-research --json.

Limitation: 60 is a target, not a guarantee. If coverage falls short, the skill reports the gap rather than padding with weak sources, and paywalled papers may only be reviewed at abstract level.

Frequently Asked Questions

Is it a systematic review tool?

No. For PRISMA-style protocols, screening, and risk of bias, it hands off to Literature Review.

Does it still use Perplexity Sonar?

Only when selected explicitly or enabled with --fallback-perplexity.

Alternatives

• Perplexity AI: interactive cited answers, without a packet or evidence matrix.
• Firecrawl: web crawling and extraction for your own pipeline.
• NotebookLM: question answering over sources you already have.

Conclusion

Research Lookup fits researchers who want a large, auditable evidence packet generated from inside Claude Code and are comfortable paying for Parallel queries. Treat returned web content as data to verify, not as instructions. More research tools are in the skills hub.