vibe coding for biology

Describe the experiment in English. Get primers, assembly plans, robot code and a reading of your plate — then every result, including the ones that failed, is committed to an immutable record you can trace back from a sequenced barcode.

What actually happens when you hit run

Design

Real calculations, not guesses

Nearest-neighbour melting temperatures from SantaLucia's 1998 unified parameters. Golden Gate fusion sites screened for palindromes, reverse-complement pairs and single-mismatch neighbours. A fluorescent protein table with published brightness and photostability.

Execute

Robot code that is checked first

Protocols emitted and validated before you load a tip: every volume against the fitted pipette's range, no deck slot booked twice, enough tips on the deck, trash left clear. It also writes down the steps the robot cannot do, because most of a protocol is still hands.

The same protocol comes out for an Opentrons OT-2, a Tecan Freedom EVO or a Hamilton STARlet. A Tecan does not require EVOware: via PyLabRobot it takes the same firmware commands the vendor's own software sends.

Record

The failures are the point

Paste the plate reader's export, mess and all. Every well is scored by median and MAD and every well is committed — the duds alongside the hits. That is the difference between "nobody tried this" and "somebody tried this and it did not work".

Why a box and not a chatbot

This box is a router, not a language model. It reads your line, decides which real tool you are asking for, runs it and shows you what came back. Nothing on this page is generated prose.

The language understanding belongs in the agent you already use. Every tool behind this box is also exposed over MCP, so Claude, Cursor or anything else that speaks the protocol can design a primer, plan an assembly, write the robot code and commit the result on your behalf, in whatever conversation you were already having.

{
  "mcpServers": {
    "sudo.bio": {
      "command": "bio",
      "args": [
        "mcp"
      ]
    }
  }
}

What it does not do, so you are not surprised

It does not predict which mutation will work. For brightness it will usually tell you not to run the screen and to swap in a published protein instead, which is cheaper and almost always better.
Tecan and Hamilton code is emitted at firmware level, which the manufacturers do not endorse and which may affect your warranty. The protocols are also designed on an OT-2 deck, so moving them needs labware substitutions — each one is listed with the script rather than hidden. For hardware PyLabRobot has no backend for, such as a Tecan Fluent, you get the transferable plan and an honest refusal to fake a script.
Validated is not verified. Run generated protocols through opentrons_simulate before the deck is loaded.
There is no cloud and no lock-in. The store is a real bare git repository on disk; git log works on it directly and you can walk away with everything.

Browse the repositories, or paste a sequenced read into the provenance trace to find out what it is.