Check the antibodies in a paper — with your own AI
Connect the Only Good Antibodies database to Claude, ChatGPT, or any MCP-capable
assistant. Then, when you ask it to read or summarise a manuscript, preprint, or
methods section, it checks every antibody the paper cites against independent,
knockout-controlled characterisation data from YCharOS (produced to community
consensus protocols) — and tells you how each one performed, application by
application.
⚗️ This is an early prototype. It works today and we use it ourselves,
but the tools, wording, and results may change as we improve it. It reports only what
YCharOS has independently characterised with knockout controls (to community consensus
protocols) — a research aid, not a substitute for reading the paper. Feedback is very
welcome.
The connector URL
This is the OGA database connector — a standard remote MCP server.
Add it once in your assistant's settings and it runs on your own usage.
https://oga-mcp.onrender.com/readonly/mcp
How to check a paper
Once the database is connected (steps below), you drive it in plain language:
Give your AI a paper. Paste the text or methods section, attach a
PDF, or share a link.
Ask it to read, summarise, or review it — or just say
“check the antibodies in this paper against OGA.” Your assistant runs OGA's
manuscript scan on the text for you; you don't have to list the reagents yourself.
Read the verdict. Every antibody it can identify (by catalogue
number or RRID) comes back grouped as recommended / not recommended / not
tested / not in the dataset, each with its per-application result
(WB / IP / IF / FC), RRID, and report DOI. Genes named in the paper are checked too.
Try this:“Summarise this preprint, then use
the OGA tools to check whether the antibodies it uses are knockout-validated: [paste
the text or a link].”
What it does
Scans the whole manuscript. It pulls every catalogue number,
RRID, and gene out of the text and checks each against YCharOS's independent,
knockout-controlled characterisation data.
Verdicts per application. A western-blot pass tells you nothing
about IF, IP, or FC — the scan keeps each application separate so nothing is
over-generalised.
Grounded, never guessed. It reports only what YCharOS has
independently characterised. “Not in the dataset” means untested — not that an
antibody is unreliable.
Add it to your assistant
The OGA database is a custom connector (MCP server). Add it once in
your assistant's settings, sign in when prompted, and it appears as a set of OGA data
tools you can use in any chat.
Claude (claude.ai)
Open Settings → Connectors.
Click Add custom connector and paste the URL above.
Sign in when prompted (this just authorises your connection).
In any chat, paste a paper and ask: “Summarise this and use the OGA tools
to check its antibodies.”
ChatGPT
Open Settings → Connectors (or Apps & Connectors).
Choose Add / Create connector and paste the URL above.
Sign in when prompted to authorise the connection.
Enable it in a chat and ask the same question.
Any MCP-capable assistant works. The OGA database is a standard
remote MCP server, so it also connects to Claude Code, Claude Desktop, and other MCP
clients — paste the same URL wherever you add connectors.
Not ready to connect an assistant? You can still use OGA in your browser:
browse any gene's independently validated antibodies across the site, or use the
Selection Tool to pick an antibody and its
controls for your target — no connector needed.
Then keep going
Choosing or validating an antibody for your own work? Our free tools help you plan it,
run the right controls, and keep a record:
