Claude Cowork plugins for universities and research offices
A Claude Cowork plugin is a package of skills, connectors and sub-agents that you install once to shape how Claude works for one job. For a university or research office, the useful split is between plugins that fetch and analyse research, and a thinking-partner plugin that works on the question before any of that. This guide covers both, and how to vet either.
Updated
What is a Claude Cowork plugin?
A Claude Cowork plugin bundles skills, connectors and sub-agents into one installable package, so Claude starts a task already set up for a role. Anthropic makes plugins available on every paid Claude plan, and the skills inside a plugin work in web chat, the Claude Desktop Chat tab and Cowork.
DEMONSTRATEDAccording to Anthropic’s help center article on using plugins in Claude, a Claude Cowork plugin is installed from the Customize menu, under the Plugins tab. The skills a plugin carries show up when you type a slash. Hooks and sub-agents run in Cowork and Claude Code, which is why the full behaviour of a plugin appears there and a lighter version appears in chat.
In Claude Code the same idea holds. The Claude Code plugins overview describes a plugin as one directory installed as a unit, holding skills, agents, hooks and MCP servers, installed with the /plugin command. So a plugin is a portable folder of instructions and connections, not a separate app.
Which Claude Cowork plugins fit research work?
Research work needs two kinds of Claude Cowork plugin. Research-tools plugins fetch and analyse: literature, datasets, preclinical databases. A thinking-partner plugin works earlier, on what the question actually is and which method should run next. Most offices need the first kind daily and the second kind at every decision.
Anthropic publishes its own plugins in the open knowledge-work-plugins repository, including a bio-research plugin that connects Claude to preclinical research tools and databases, and a data plugin for exploring datasets. Those are research-tools plugins. The table below sorts the jobs a research office actually has by the kind of plugin that serves them.
| The job | Kind of plugin | What it hands back |
|---|---|---|
| Search literature or a preclinical database | Research tools | Papers, records, summaries |
| Explore a dataset someone dropped in a folder | Data analysis | Column summaries, anomalies, charts |
| Decide which question deserves the next grant cycle | Thinking partner | A classified problem and confirmed claims |
| Look across disclosures for a pairing nobody filed together | Thinking partner | Cross-field idea pairs, each with a confidence label |
| Teach a cohort one method for problem finding | Thinking partner | A room per student with reasoning you can read |
The thinking-partner rows matter most when the problem is not yet defined. If you have seen a team polish an answer to the wrong question, the essay on why most teams solve the wrong problem shows the pattern.
How should a research office vet a Claude plugin?
Vet a Claude plugin on five points: where it runs, what leaves the machine, who can approve what it writes, whether your admin can control it, and who published it. Anthropic’s own documentation supplies most of the questions, because local plugins run with the same permissions as any program on your computer.
Anthropic’s plugin guidance explains that local MCP servers run with local program permissions. Enterprise administrators can restrict plugins or disable local servers. Check your organization’s policy before installing one.
| Question to ask | Why it matters | Where to check |
|---|---|---|
| Does it run a local server or call a remote one? | Local execution and remote connector requests have different access and data boundaries | The plugin’s manifest and Anthropic’s connector docs |
| What text leaves the machine, and to whom? | Unpublished results and disclosures are the asset | The publisher’s privacy page |
| Does a person approve what it writes down? | A committee needs to see who confirmed a claim | A trial run on a real case |
| Can your admin restrict or distribute it? | Enterprise admins can limit plugins and turn off local servers | Your Claude organization settings |
| Who published it, and is the code public? | Trust is the install decision | The repository or package page |
For m:os, the answers are on the privacy page and in the section below. Anthropic also documents that on the Enterprise plan, plugins can be scanned for malicious content when installed, and that admins can run private plugin marketplaces to distribute approved plugins across an organization.
What is a thinking-partner plugin?
MindrianOS helps you examine a hard problem or a load-bearing hypothesis. Theo contributes questions, methods and frameworks. The DataRoom holds the evolving context. A new relationship may reveal an opportunity, but you decide what the evidence supports and what to test next.
A useful starting point is a hypothesis, the decision it supports and one real source. Form a problem statement, break it into smaller questions and examine the relationships between them. The working-session guide shows this progression; the command reference describes individual operations.
The first DataRoom walkthrough asks you to review the proposed structure before approving it. Sections should follow the work rather than force every research question into a venture template.
THESISOur argument is that better questions can change which research deserves the next test. It is not a measured claim that using MindrianOS produces breakthroughs. Read how a scientific eureka happens for an example of a changed relationship creating a new possibility.
What does m:os send off your machine?
The plugin keeps DataRoom material locally. Its Theo method-routing boundary is designed around the problem shape, including the problem type and method names. This is separate from the context processed by Claude and any other connected tools. Review the whole workflow before sharing unpublished research.
The Theo connection guide explains the specific boundary and diagnostics. The privacy page describes website processing. Local storage alone does not mean that an AI session is entirely local.
Does m:os work in Claude Code and Claude Desktop?
Claude Code is the documented installation path for MindrianOS. Desktop and Cowork require checking the host’s plugin capabilities, permissions and access to the project. A method connector is only one part of the system; it does not establish full runtime parity or automatically synchronize a DataRoom.
Follow installation, then verification. Use the surface guide to distinguish local plugin operations from Theo method guidance.
What should a pilot establish?
A useful pilot should establish that people can install MindrianOS, bring appropriate context into a DataRoom and explain a change in their understanding. A proposed opportunity remains unvalidated until evidence supports it. Agree on the next evidence-producing step and who has authority to approve it.
Use the institutional evaluation guide to frame a bounded case. The research page focuses on load-bearing hypotheses. Product developments can be followed in the m:os letter.
Frequently asked questions
- Do Claude Cowork plugins need a paid Claude plan?
- Yes. Anthropic's help center says plugins are available on all paid plans: Pro, Max, Team and Enterprise. The m:os plugin itself is free; it runs inside the Claude plan you already have.
- Can a university admin control which plugins staff install?
- On the Enterprise plan, an admin may restrict which plugins people can install or turn off local MCP servers entirely, and owners can distribute plugins through an organization marketplace. Check with whoever runs your Claude organization before a pilot.
- Does m:os send our research to a server?
- The plugin stores DataRoom material locally and limits its Theo method-routing requests to problem shapes. Claude host processing and other connected tools have separate data policies. Review every data path before using unpublished research.
- Is m:os a Claude Code plugin or a Cowork plugin?
- Claude Code is the documented installation and verification path. Desktop and Cowork capabilities depend on the host, permissions and plugin configuration; a Theo connector alone does not install the complete local runtime.
- What does m:os do that a research-tools plugin does not?
- A research-tools plugin fetches and analyses: literature, datasets, databases. m:os works one step earlier, on the question itself. Theo contributes questions and methods. The DataRoom keeps the investigation’s context, and you judge the evidence and choose what to commit.
A note on evidence: platform facts on this page come from Anthropic’s help center, Claude Code documentation, the knowledge-work-plugins repository and Anthropic’s product blog, each linked where used. Product workflow guidance links to the installation and DataRoom documentation. THESIS identifies our argument, not a measured outcome.
What does this change about your problem?
Bring a hard problem or a load-bearing hypothesis into a DataRoom. Connect the context, question the assumptions, and choose what to test next.