Use Biosimulant from Claude Code or Codex

Two lines to connect. After that you just ask. Your agent finds the model, works out what it is, and runs it when you say so.

Free during the beta. Nothing runs without your say-so.

Connect it

Sign in once, from inside the agent. It sends your authorization on every call after that. The beta is free and there is no balance to top up.

$ claude mcp add --transport http biosimulant https://api.biosimulant.com/mcp
/mcp

Run this inside Claude Code to sign in.

What you are agreeing to

  • Read the Hub and the biological sources we support
  • Make and edit private workspaces of your own
  • Start runs, after you have approved them
  • Publish releases, after you have approved them
  • Stay connected until you disconnect it

What you would actually ask it

Three things people do on the first day. The replies below come from a real session, shortened to fit.

Find a model and work out what it is

Searches the Hub, pins the exact release, then goes and reads the paper behind it.

Nobody asked it for that last line. It tells you what the package has not been checked for.

Find the Bexkens 2024 S. mansoni glycolysis lab on the Hub and explain what it models.

Founddemi/bexkens2024-schistosoma-mansoni-glycolysis@1.0.0

SHA-256ea0fed3fd767794455b514a2d8a5d74200c595b38850b3e01b34367158ac4843

A kinetic model of glycolysis in S. mansoni, focused on how allosteric regulation of lactate dehydrogenase affects glycolytic behaviour.

Paper10.1016/j.ijpara.2024.03.005 · PMID 38492780

Qualification: schema_valid: true, but runtime_valid: false. Treat it as a package with a valid schema, not a domain-validated reproduction.

Fork it and run it

Copies the release into a private workspace of yours, gets a compute session ready, and stops.

It can set the whole thing up. It cannot start the run.

Fork it into a workspace and run it with the default inputs.

Waiting for your approval

Once you allow it:

Runcompleted · duration 10.0 · step 1.0

Resultsresults/default_run_results.json

Of the selected observables, AMP moved most across the window. ATP, ADP, pyruvate, NAD and NADH were reported too.

Check two models fit together

Reads both ports and compares the profile each one declares.

Two ports connect when both sides name the same profile. Otherwise it stops early, rather than giving you a number that means nothing.

Can I feed this protein sequence into the Boltz affinity model?

protein_sequence declares protein.sequence/v1. So does the port you want to wire it to, so the connection holds.

ligand_smiles declares chemical.smiles/v1. Wiring a sequence into that one fails the check before anything runs.

Nothing runs until you say so.

Every tool that would change something, spend compute or publish comes in two halves. The agent can only reach the first one on its own.

  1. prepare

    The agent works out exactly what it wants to do. Nothing changes yet.

  2. You read it

    You get the exact plan and the digest it is pinned to, not a summary of one.

  3. You approve

    Or you do not, and it stops there.

  4. apply

    Only now does anything run, save or publish.

If approving one step at a time gets tiring, hand over a grant instead. It covers a set of actions you name, it expires on its own, and you can pull it whenever you want.

What it can reach

Find
Search the Hub, pin an exact release by digest, and pull records from the biological sources we support.
Build
Make a private workspace, fork a release into it, and change models and compositions inside it.
Run
Prepare a run, start it once you approve, watch it, cancel it, and fetch what it produced.
Prove
Read the signed Passport for any run: the digests, the runtime version and the inputs that made the result.
Publish
Release your work to the Hub, privately or publicly, and only when you have said so.

Keeping files off the cloud

Use the local CLI when your files have to stay on your own machine. Local work is not saved as a revision and gets no managed run or Passport.

Install the CLI

Skills

Two specification documents, for writing formal model requirements and a technical plan before you build. They are instructions, not tools: no extra access, no extra permissions. Most people never need them.

Download skills

The endpoint is https://api.biosimulant.com/mcp. Something not working? Read the Gateway guide (opens in a new tab)