bgpt-paper-search

Trigger phrases

Phrases that activate this skill when typed to Claude Code:

• search BGPT for papers
• find experimental data on
• get study details for
• extract methods from papers on

What it does

bgpt-paper-search is a Claude Code skill from K-Dense AI’s scientific-agent-skills repo. It turns Claude into a structured data extractor for scientific literature by connecting to the BGPT MCP server at bgpt.pro/mcp. Unlike abstract-only searches, BGPT returns 25+ structured fields per paper extracted from full text: methods, results, sample sizes, effect sizes, quality scores, and conclusions.

A session produces a structured dataset of study characteristics — the kind of extraction table that takes hours to build manually for a systematic review or evidence synthesis project.

When to use it

Reach for it when:

• You’re building an evidence synthesis and need structured methods/results tables, not just abstracts
• You want to compare sample sizes, outcome measures, or statistical approaches across a set of studies on the same topic
• You need quality scores or risk-of-bias indicators as part of a systematic review workflow

When not to reach for it:

• Simple abstract retrieval or DOI lookup — paper-lookup is faster and doesn’t require the MCP server
• Broad exploratory searches where you don’t yet know what structured fields you need

Install

Copy the SKILL.md from K-Dense AI’s bgpt-paper-search folder into .claude/skills/bgpt-paper-search/ in your project. This skill requires the BGPT MCP server to be configured — see bgpt.pro/mcp for setup instructions.

Trigger phrases: “search BGPT for papers”, “find experimental data on”, “extract methods from papers on”.

What a session looks like

A typical session has three phases:

1. Query design. Specify the topic, intervention, population, or outcome of interest. The more specific the query, the more useful the structured extraction.
2. BGPT retrieval. Claude calls the BGPT MCP server, which returns papers with full structured extraction — each paper as a record with 25+ fields rather than a flat text block.
3. Synthesis output. Claude organizes the extracted records into a comparison table or narrative synthesis, with sample sizes, methods, and findings side by side.

Receipts

Where it works well:

• Biomedical RCTs and observational studies where BGPT’s extraction models are trained — structured fields are reliably populated
• Evidence synthesis projects where manual extraction would take a research assistant days

Where it backfires:

• Basic science papers and preprints outside BGPT’s training distribution may return incomplete field coverage
• The MCP server adds a dependency that must be set up separately — not a zero-configuration skill

Pattern that works: use paper-lookup to get a candidate set first, then pass the DOI list to bgpt-paper-search for structured extraction rather than querying cold.

Source and attribution

Skill authored by the BGPT team. The canonical SKILL.md lives in the bgpt-paper-search folder of K-Dense AI’s scientific-agent-skills repository.

License: MIT. Install, adapt, and redistribute with attribution preserved.

This page documents the skill from a practitioner’s perspective. For the formal spec and any updates, defer to the source repo.