Claude SandboxAQ connectorWhat Claude can do in your SandboxAQ account.
The Claude SandboxAQ connector exposes 7 tools: 6 read, 0 write, 1 undocumented. They run the publisher's scientific models and bring the results back into the chat. Below: what Claude computes, where it gets stuck, and who can cut access.
Verified Trustpilot reviews · AI, automation & growth agency
What changes for an R&D team
Pre-screen a catalyst. For a given material, adsorbate and facet, AQCat tries several placements and returns the lowest binding energy, with the convergence status of each placement.
Rank a compound library. Against one target, AqpotencyScreen scores a library of SMILES strings (the text notation for a molecule) for virtual screening, up to 10,000 compounds.
Catch off-target risk. For one ligand, AqpotencyScan runs it against a chosen target panel, then AqpotencySelectivity puts a number on the gap between the intended target and a list of others.
What Claude doesn't do here: nothing fires on its own, every run follows a question you ask. You need an active SandboxAQ account, and the publisher bills each model by usage, separately from Claude. The docs also warn that the language model may answer from memory instead of calling the tool. For scheduled batch jobs, you want an automation tool; the Integrations page covers that side, and the other connectors sit under Claude connectors.
The vocabulary in one minute
Five words worth knowing before you connect SandboxAQ.
- Connector
- The link you set up once between Claude and an account you already have, so Claude can work on it while it answers you.
- Tool
- One named action a connector opens to Claude. Claude picks the ones it needs on its own; the directory lists them by name.
- Authorization
- The service's own sign-in screen, where you hand Claude the access it will use. Granted once per person, and you can take it back.
- Approval
- The confirmation Claude waits for before finishing something that changes your account, shown in the chat right when it matters.
- MCP
- The common standard behind connectors: it's what lets an assistant like Claude hold a conversation with an outside service.
Connect SandboxAQ to Claude in three steps
- 01
Find SandboxAQ
In Claude's settings, open Customize, then Connectors, and spot SandboxAQ in the list. On a Team or Enterprise workspace, an Owner or Primary Owner has to enable it before each member can sign in.
- 02
Start the connection
Click Connect on its row, then sign in through the window the service opens itself. If the link breaks one day, hit Disconnect and connect again: that's the intended fix.
- 03
Read the authorization screen
Read the authorization screen before you approve. It belongs to the service, not to Claude, and it decides what the access covers. Per SandboxAQ, the account you sign in with is the SandboxAQ MCP Server account created on its platform.
The 7 tools, grouped by what they do
SandboxAQ gives Claude 7 tools: 6 that read your account, 0 that change something in it, and 1 no official source describes.
Five model tools, one tool that tracks runs, and one name nothing documents. Names stay exactly as Claude shows them.
- 6 read
- 1 not documented
What Claude computes and checks (6)
6 toolsThe publisher's docs describe these as inference operations with no writes.
AQCat
Runs an automated adsorption screen for one bulk material, one adsorbate and one crystal facet, then reports the lowest binding energy across the placements it tried, with convergence for each one.
AqpotencyPotency
Predicts the pIC50 of a protein-ligand pair from a UniProt accession and a SMILES string. It's the simplest case: one protein, one molecule, one predicted potency value.
AqpotencyScreen
Scores an uploaded SMILES library against a single target for virtual screening, up to 10,000 compounds.
AqpotencyScan
Runs one ligand against a curated target panel (human kinome, Bowes safety panel or proteome) for reverse screening and off-target triage.
AqpotencySelectivity
Computes on-target potency against a named list of off-targets, then the fold-selectivity between them, for a single ligand at a time.
Check Job Status
Tells where an AQPotency run stands, since those four tools first answer with a job ID. Per the publisher, the client handles the polling loop.
Undocumented (1)
1 toolThe directory publishes this name, and no official text says what it does. We won't guess from the label.
Check Job Results
This label sits in the directory's tool list, and no official source describes it. The publisher's docs do mention fetching a job's results, but under another name: nothing confirms it's this tool.
What Claude asks before it acts
By default, Claude stops and asks for your go-ahead before any action it takes on an account for you. The request shows up in the chat, right when it matters.
No source describes a confirmation specific to any SandboxAQ tool, so that default rule is what applies. On a Team or Enterprise workspace, owners decide whether a member can let certain actions through without being asked again, and they can limit what a connector may do for the whole organization. Nobody overrides that from their own account. Finally, Claude works with your rights and nothing more.
Which plans include it
Out of 819 sheets in the official directory, none shows availability by plan. Connector by connector, the answer isn't published anywhere: it's a real gap in the catalogue.
The general rule is published: remote connectors are open to all users on Claude, Cowork, Claude Desktop and mobile. On Team and Enterprise, an Owner or Primary Owner turns the connector on for the organization before each member can connect. For the current status, the SandboxAQ sheet in the official directory is where to look.
Where SandboxAQ stops inside Claude
A connector isn't an automation. Claude calls these tools while it answers you: no run starts by itself when a new molecule lands in your database.
Only the publisher documents this connector: neither Claude's help center nor Anthropic's docs mention it. By default its sheet exposes reads only, which is an observed floor, not a guarantee. The directory's partner badge isn't a security audit, and Anthropic says so on every sheet: it doesn't choose the tools a publisher exposes and doesn't vouch for how they behave. Connect it only if you trust SandboxAQ itself.
Need help connecting SandboxAQ to Claude?
A person reads every message.