Connect Mass Dynamics to the AI tool you already use. Your agent works on your live workspace, with your data where it is, under the same rigour as the work you do by hand.
Nothing to export. Nothing to install. No terminal.
Ask in plain language. Your agent does the work against your workspace and tells you what actually happened.
Hand it the file and it goes in, validated on the way. Problems surface as you upload rather than an hour later.
Group the samples, name the conditions, set the metadata. All by asking, against the same workspace you use in the browser.
Kick off an analysis, then ask where it got to. No hunting through a job list to find out whether it finished.
Your agent reads the actual error rather than a status code, so it can usually tell you what went wrong and fix it.
You need a Mass Dynamics account. Everything else happens in your own tool.
https://app.massdynamics.com/mcp
In your tool's connector settings, add the URL above as a custom connector. Exact wording varies by tool.
You'll be sent to Mass Dynamics to authorise the connection with your normal login. No API key to copy or store.
"List my datasets" is a good first move. If that comes back, you're connected and can start uploading.
The tools we support today. AI clients change often and without notice, so we keep this list current as they do. If a connection stops working, contact support.
| Tool | Status | Notes |
|---|---|---|
| Claude Desktop | Supported | Add as a connector. |
| Claude Code | Supported | Add the server URL and confirm when prompted. |
| Codex | Supported | No additional configuration. |
| ChatGPT Desktop | Extra setup | Requires developer mode enabled, and the connector added once via ChatGPT in a browser before the desktop app will use it. |
| ChatGPT on the web | Partial | Datasets, workflows and results are supported. Experiment file upload is not. Use the desktop app to upload. |
Not listed does not mean unsupported. Contact support with the tool you use and we will work through it with you.
Common questions, with answers you can forward.
Your agent connects to your existing Mass Dynamics workspace. No copy of your data is created and the agent holds no separate store.
Authentication. Access is authorised over OAuth 2.0. You sign in with your normal Mass Dynamics credentials and the agent is issued a token for your account. It operates with your permissions and can reach only what your account can reach. No API key is created or stored on your machine.
Uploads. Files sent through an agent take the same path as files uploaded through the browser: the same storage and the same processing pipeline, into the same workspace. Existing data is not moved or duplicated.
What the connection carries. The agent calls specific operations, such as listing datasets, fetching a result, starting a workflow or reading a job status, and receives the response to each call.
Further detail is in our security and compliance documentation.
Two systems, two different answers, and the second one isn't ours to give.
On our side. Your agent asks for specific things: a list of datasets, a set of results, an error message. It gets back what it asked for. Your data stays in your Mass Dynamics workspace, and what leaves is whatever the agent requested to answer your question.
On your AI provider's side. Whatever your agent reads passes through the model in order to answer you, and what happens to it there is governed by your agreement with Anthropic, OpenAI or whoever you use. That's the part worth checking, and the answer usually differs between a personal account and a team or enterprise plan. Enterprise and team plans generally carry stronger retention and training commitments than individual ones.
If your organisation has an enterprise agreement with an AI provider, connecting through that rather than a personal account is the stronger position, and an administrator can deploy the connector across your whole organisation. Our security and compliance documentation covers the Mass Dynamics side. For anything further, contact support.
No. That's what this path is for. You add a connector once, then ask for what you need in plain language. If you would rather drive it programmatically, the API is the better tool for that job.
This is the production version of what the Agent Pack was testing. It is hosted, so there is nothing to install, and the error handling that used to leave you stuck now lives in the server. The Agent Pack itself has been retired.
If your old setup has stopped working it is inert and you can leave it. Remove it if your agent starts offering duplicate Mass Dynamics tools and picking the wrong one.
Removing an Agent Pack setup has the steps, including retiring the API token the old installer saved to your disk. Deleting the files does not revoke that token.
Both, usually. The agent for exploration and the day-to-day. The API for anything that has to run identically every time, including instrument and vendor output formats. Same platform underneath either way.
Yes. We do not limit which tools you connect or how many. An agent is another interface onto the same workspace, in the same way the web app and the API are.
Each connection authenticates separately under your own login, and all of them act on the same workspace. Work done through one tool is there in the others and in the browser. The same holds across a team: each person connects with their own account and their own permissions.
The one thing to avoid is connecting two copies of the Mass Dynamics server at once. Your agent then sees duplicate tools and may call the wrong one. That is why an older Agent Pack setup should be removed before you connect.
Yes, and for larger organisations that's usually the better route. An administrator deploys the connector across your Anthropic or OpenAI corporate account rather than everyone setting it up individually. Contact support and we will work through it with your administrators.
Connecting your agent doesn't change your plan.
The list covers the tools we support today. It is not a limit on what can connect, and an unlisted tool that speaks MCP will often work already. Contact support with the tool you use and we will confirm it and add it to the list.
Thank you for testing it. What you found is why this version is better: it's hosted, so there's nothing to install, and the error handling that used to leave you stuck now lives in the server. The Agent Pack has done its job and we've retired it.
You may not need to do anything. If the old setup has stopped working, it's inert and you can leave it. Clean it up if your agent starts offering you two sets of Mass Dynamics tools and picking the wrong one.
Please audit my Mass Dynamics setup without changing anything. Look for any MCP server entries mentioning mass dynamics and whether each points at a local install or a hosted URL, any Mass Dynamics skills or plugins, and any config or .env file holding a Mass Dynamics API token. Tell me what you find and where each one lives. Don't remove or edit anything.
md-python folder in your home directory and an entry in your Claude Desktop config file.Not sure what you're looking at? Send the audit to support and we will tell you what is safe to remove.
Most people are connected in a couple of minutes using the steps above. If yours isn't one of those - an unusual setup, a tool that isn't on the list yet, or something that just won't behave - tell us and we'll work it out with you.
No Mass Dynamics account yet? Say so below and we'll sort that out first.
Which tool you're using and where it got stuck. That's usually enough for us to know the answer. Takes about two minutes.