mcpField

bgpt-mcp

github.com/connerlambdenProductivity61 installs
Repository

About bgpt-mcp

BGPT is a remote MCP server for searching scientific papers and retrieving structured study details, including methods, results, limitations, conflicts of interest, and quality scores. It is intended for literature searches and evidence-focused appraisal.

Key features

  1. 01Search papers with natural-language queries
  2. 02Retrieve methods and experimental results
  3. 03Surface limitations and study blindspots
  4. 04Return conflict-of-interest and funding details
  5. 05Provide calibrated quality scores

Use cases

  1. 01Find research on a scientific topic
  2. 02Appraise a paper's reliability and limitations
  3. 03Compare reproducibility across studies
  4. 04Review sample sizes and study populations

MCP support

BGPT exposes a single search_papers tool to MCP clients. Results include bibliographic details and structured fields covering study methods, findings, limitations, and quality signals.

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/connerlambden/bgpt-mcp --skill bgpt-mcp

After installing, type/bgpt-mcpin Claude Code to confirm it loaded.

Details

From the project README

BGPT MCP — Scientific Paper Search with Structured Critical Appraisal

BGPT is a remote Model Context Protocol (MCP) server that searches a database of scientific papers built from full-text studies. Unlike typical search tools that return titles and abstracts, BGPT extracts raw experimental data along with 25+ structured fields per paper.

When to Use

Use BGPT whenever the user asks to:

  • Find scientific papers on a topic
  • Evaluate research quality, reproducibility, or COI
  • Surface limitations, blindspots, or falsifiability criteria of studies
  • Retrieve methods, sample sizes, populations, or experimental techniques
  • Build an evidence-grounded answer that requires citing primary literature
  • Compare evidence across multiple studies with quality signals

Setup

Add BGPT to the MCP client config — no API key needed for the free tier (50 free results, then $0.02/result).

Claude Desktop / Cursor / any MCP client (SSE):

{
  "mcpServers": {
    "bgpt": {
      "url": "https://bgpt.pro/mcp/sse"
    }
  }
}

Streamable HTTP (alternative): https://bgpt.pro/mcp/stream

Tool: search_papers

The single tool exposed by BGPT.

Input:

  • query (string, required) — natural-language description of what to search for. Examples: "CRISPR base editing in vivo delivery 2024", "semaglutide cardiovascular outcomes", "transformer attention sparse activations".

Output (per paper):

Each result is a paper with these structured fields:

  • title, authors, journal, year, doi, url
  • abstract
  • methods_and_experimental_techniques
  • results_and_conclusions
  • sample_size_and_population
  • study_context
  • paper_limitations_and_biases
  • study_blindspots — confounders the authors did not examine
  • how_to_falsify — what evidence would disprove the paper's claims
  • conflict_of_interest
  • funding_json
  • data_availability_statements
  • code_and_data_links
  • quality_scores — calibrated 0-10 scores for: scientific rigor, novelty, generality, usefulness, reproducibility, depth

Best Practices

  1. Cite the structured fields, not just the abstract. When summarizing a paper, surface the paper_limitations_and_biases, study_blindspots, and conflict_of_interest fields in the response so the user gets calibrated context.
  2. Use quality_scores.reproducibility to triage. When multiple papers conflict, prefer the higher reproducibility score and note the difference.
  3. For systematic reviews and clinical questions, leverage sample_size_and_population to filter underpowered studies.
  4. For methods-heavy questions (replication, technique comparison), focus on methods_and_experimental_techniques.
  5. Always show how_to_falsify when the user asks "is X true?" — the falsification criterion is the test the claim should pass.
  6. Always disclose conflict_of_interest when summarizing pharma, nutrition, or industry-funded research.

Example Workflows

Evidence-graded literature search

User: "What's the evidence for time-restricted eating on cardiovascular outcomes?"

Use search_papers("time-restricted eating cardiovascular outcomes RCT"). From results, return a comparative table that includes for each paper: sample size, primary outcome, reproducibility score, and disclosed COI/funding.

Critical appraisal

User: "Is paper X reliable?"

Search for the paper title. Surface paper_limitations_and_biases, study_blindspots, how_to_falsify, and quality_scores. State the falsification criterion explicitly.

Replicability check

User: "Has this finding replicated?"

Search the original finding, then search for follow-up replications. Compare quality_scores.reproducibility across studies and surface methodological differences from methods_and_experimental_techniques.

Pricing

  • 50 free searches per network (no API key needed)
  • $0.02 per result thereafter (API key from https://bgpt.pro/mcp)

Links

Comments

Markdown supported