search_articlesLABSCO SUMMARY
A literature search that ends with the abstracts, not a list of links.
Search and retrieval are both here, and retrieval is batched — so a query becomes a set of abstracts an assistant can actually read and compare, rather than a list of PMIDs somebody has to open one at a time.
What it is
An MCP server over PubMed's E-utilities API, giving an assistant search across the biomedical literature plus retrieval of article metadata, abstracts and available content.
What you get
- Search across PubMed with flexible query terms, including boolean expressions, with result count and sort order under your control
- Full retrieval for an article: metadata, abstract and whatever content the record exposes
- Batch download, so a reading list comes back in one request rather than one call per paper
- Document summaries with metadata, when the full record is more than the question needs
- Output in XML, JSON or plain text
- Rate limiting applied automatically, so a batch does not trip NCBI's per-second limits
Requirements
Python in a virtual environment with the packages in requirements.txt; a setup script does this in one step. No key is required, but an NCBI API key raises the rate limit from 3 to 10 requests per second, and NCBI asks for a contact email — both go in a .env file.
Setup effort
Build from source — git clone https://github.com/aeghnnsw/pubmed-mcp.git
USE CASES
01
Finding out what has been published on a question
Run the query with the boolean terms you would use on PubMed itself, and control how many results come back and in what order.
02
Reading the abstracts rather than the titles
Pull the full records for the hits that look relevant, so the screening decision is made on the abstract.
03
Working through a reading list in one pass
Download several articles in a single request instead of a call per paper, which is what makes screening a reading list practical.
04
Checking a citation before you use it
Fetch the summary for a specific record and confirm the authors, journal and year against the source.
05
Feeding structured records into your own pipeline
Take the output as XML or JSON when the results are going into a script rather than into a conversation.
4 tools
Listed in project docs. Your client may show a different list.
download_articledownload_articles_batchget_article_summariesBEFORE YOU INSTALL
What to know first
Tools you get
4 tools named in the project's documentation.
Where it runs
The documented client runs the server on your machine.
INSTALLATION
Connect your client
You clone this one yourself.
Server setup
TERMINAL
git clone https://github.com/aeghnnsw/pubmed-mcp.git
cd pubmed-mcp
./setup.sh
cp .env.example .envEdit .env with your NCBI API key and email (optional but recommended)
Add the Claude Desktop block on this page to your Claude Desktop config, using the full path to the folder you cloned
MCP Clients
Claude Desktop
CLAUDE DESKTOP · JSON
{
"mcpServers": {
"pubmed": {
"command": "<YOUR_PATH_TO_PUBMED_MCP>/venv/bin/python",
"args": [
"<YOUR_PATH_TO_PUBMED_MCP>/server.py"
]
}
}
}
