BGPT MCP — Search scientific papers with full-text experimental data

[connerlambden]

Pre-submission Checklist

• This post follows the MCP community guidelines

What would you like to share?

What I built

BGPT MCP API — a hosted MCP server that lets AI assistants search scientific papers using data extracted from full-text studies, not just titles and abstracts.

It exposes one tool: search_papers, which returns 25+ structured fields per paper including methods, results, conclusions, sample sizes, limitations, conflicts of interest, and quality scores.

How it works

BGPT runs as a remote SSE server at https://bgpt.pro/mcp/sse. Any MCP client can connect with a simple config:

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

No Docker, no local installation, no API key needed for the free tier (50 searches).

What makes it different

Most paper search tools return titles and abstracts. BGPT extracts data from the full text of published studies, so your AI can reason over:

• How the study was actually run (methods, experimental techniques)
• What was measured (results, effect sizes, statistical outcomes)
• What might be wrong with it (limitations, blind spots, conflicts of interest)
• How rigorous it is (quality scores for scientific rigor, reproducibility, etc.)

Example prompts

• "Find RCTs on GLP-1 agonists and long-term weight maintenance from the last 3 years."
• "Search for CAR-T cell therapy trials reporting both response rates and severe adverse events."
• "What are the reported limitations in CRISPR off-target effect studies?"

Pricing

• 50 searches free (no API key)
• Then $0.01 per result returned (pay-as-you-go via Stripe)

Tips for others building MCP servers

• SSE transport makes hosting simple — no need for users to install anything locally.
• Keeping the tool surface minimal (one tool with a few parameters) makes it easy for AI assistants to use correctly.
• Structured, field-level data is much more useful to LLMs than raw text blobs.

Would love feedback from the MCP community on what fields or domains to prioritize next!

Relevant Links

• GitHub repository: https://github.com/connerlambden/bgpt-mcp
• Documentation & setup: https://bgpt.pro/mcp/
• SSE endpoint: https://bgpt.pro/mcp/sse