The copy-paste era in biotech is coming to an end
- Jun 5
- 2 min read
AI applications in biotech are increasing drastically. Models can predict protein structures, summarize papers, generate hypotheses, and help researchers navigate vast amounts of scientific literature. The individual capabilities are no longer the bottleneck.
What remains surprisingly difficult is the work between those capabilities.
It is not uncommon to be in the middle of analyzing a promising compound and fall into a familiar ritual, open ChEMBL, export a CSV, switch to UniProt to find an accession number, search PubMed for supporting literature, read through abstracts, then manually stitch everything together in a spreadsheet. Maybe AI helps summarize a paper or extract a key result along the way. Switching between tabs and windows and tools and copy pasting has become an unwanted part of the workflow.
There's a fix now. It's called MCP, and it's moving faster than most people realize.
What is MCP
Model Context Protocol is an open standard built by Anthropic but now used universally that lets AI agents connect directly to external data sources and tools. There are now thousands of MCP servers and the number is increasing everyday.
The analogy that actually clicked for me was USB-C for AI. Before USB-C, every device had its own cable. After USB-C, one port works everywhere. MCP does that for AI and data, one standard, any database, any model.
What that means in practice, instead of you being the connector between ChEMBL and UniProt and PubMed, the AI is. You ask your question once. It talks to all three sources and gets back to you with structured data. The 45-minute manual loop becomes a single conversation turn.
It is also super easy to try it
The zero-effort starting point is Claude's built-in PubMed connector. Enable it and then ask Claude something you'd normally go to PubMed for. Watch it pull live results and cite them. It's a small thing but it's the moment the concept clicks. For drug discovery, the ChEMBL MCP server is useful for compound search, bioactivity, and clinical status.
Anthropic's life sciences GitHub has a catalog of what's available if you want to explore more.
The part that's still annoying and where it's getting solved
Public database MCPs are now easily available. For your data as well, it might be changing faster than expected. Benchling just launched an MCP server, and if you work in biotech you know why that's a big deal. Benchling is where a huge chunk of the industry actually stores its experimental data. Their MCP respects your existing Benchling permissions, so your AI isn't reaching past your access controls. It means Claude can query your lab notebook entries, your assay results, your registered sequences the same way it queries PubMed. That's the gap closing. Not theoretically, not on a roadmap but in reality.
There's still real work involved for systems like clinical data warehouses, legacy instruments, anything that predates modern APIs. But start with what's already wired. The tools that eat most of your manual time PubMed, ChEMBL, ClinicalTrials, and now Benchling are ready to go.
The copy-paste ritual had a pretty good run but not for long!
References:
Edited with Claude and ChatGPT
Comments