Connect a biomedical corpus to Claude with MCP

The Model Context Protocol (MCP) lets Claude call external tools and data sources in a standard way. With medground, you give Claude a grounded biomedical literature corpus, so its answers come from real papers you can open, not from model memory. Here is how to set it up.

What MCP gives you

MCP is an open standard for connecting an assistant to tools and context. Instead of pasting papers into a chat, you run a small server that exposes retrieval, evidence summaries, and a grounding check as callable tools. Claude then uses them on demand, and every answer is anchored to a citable source.

What you need

• Python 3.11 or newer
• uv, the Python package manager
• Claude (desktop or CLI) with MCP support

Set it up

Clone the project, build a corpus, and register the server with Claude:

# clone and install
git clone https://github.com/ArnaudTurn-pro/medgroundAI
cd medgroundAI
uv sync

# build a corpus
uv run medground ingest civic

# connect it to Claude
claude mcp add medground -- uv run --directory . medground-mcp

# optional: install the /doc slash commands
./install-skills.sh

What you can do once it is connected

Ask Claude a research question and it retrieves from your corpus, drafts an answer with per-claim citations, and runs a grounding check before responding. The /doc skills add focused lenses on top:

• Summarize the evidence on a question, with each claim cited.
• Match a biomarker to curated CIViC therapy evidence, leveled A to E.
• Surface where the literature disagrees, with both sides sourced.
• Appraise the quality of a single study.

Frequently asked questions