Claude Desktop Basics: Getting the Most Out of What You Have

Who this is for: Researchers already using Claude Desktop who want to use it better — and who may not have explored Projects, custom instructions, or plugins.

The four layers of Claude Desktop

Claude Desktop has four layers of capability, from basic to more powerful:

  1. Chat — the default; one conversation at a time, no memory
  2. Custom instructions — persistent instructions that apply to every conversation
  3. Projects — organised workspaces with persistent files and context
  4. Integrations / MCP plugins — connections to external tools (Zotero, Google Drive, etc.)

Most users stay at layer 1. Layers 2 and 3 already change the experience significantly.

Layer 2: Custom instructions

Custom instructions are a block of text that is prepended to every conversation you have, automatically. Claude always reads them before responding to you.

Where to set them: Settings → Profile → Custom instructions (or similar, depending on your version)

What to put in them:

I am a historian specialising in medieval religious movements.
I prefer concise responses. Do not add disclaimers or unsolicited caveats.
When summarising academic texts, include the main argument, key evidence, and any methodological notes.
My citation style is Chicago author-date.

Effect: Claude calibrates every response to your profile. You stop getting explanations of basic concepts. You stop getting hedging language that assumes you are a general public user.

Custom instructions apply to ALL conversations. For project-specific context, use Projects (layer 3).

Layer 3: Projects

Projects are persistent workspaces. Each Project has:

When to use a Project:

Creating a Project

Left sidebar → New Project → give it a name.

Inside a Project:

What persists in a Project

Without Project With Project
Conversation history Lost when tab closes Saved
Attached files Lost after conversation Persist across conversations
Custom instructions Global only Global + Project-specific
Context from previous conversations No No (each conversation still starts fresh)

Important: Files in a Project are available in every conversation, but Claude still does not remember what you discussed in previous conversations. The files are always there; the conversation history is readable but Claude does not automatically carry it forward.

Practical Project setup for a research paper

Project: "Article on Montaillou networks"

Project instructions:
  This project is about my article on social networks in Montaillou, 1300-1320.
  Key argument: kinship ties structured heretical belief transmission.
  Main sources: Fournier Register, Duvernoy edition.
  Output style: academic English, footnote citations.

Attached files:

  - my_draft.docx          (current draft)

  - key_sources_list.md    (bibliography notes)

  - structure_outline.md   (article structure)

Now in any conversation in this Project, Claude knows the context without you re-explaining it.

Layer 4: Integrations and MCP plugins

Claude Desktop can connect to external tools via integrations (Anthropic's term) or MCP servers (the underlying protocol). These give Claude access to data and services outside the conversation.

Installing integrations: Settings → Integrations (or Connections) → browse or add

Commonly useful integrations for researchers:

Integration What it adds
Zotero Claude can search your Zotero library, retrieve metadata, suggest citations
Google Drive Claude can read documents from your Drive without downloading them
Brave Search / Web search Claude can search the internet during conversation
Filesystem Claude can read files from a specific folder on your computer

Note on the Filesystem integration: This is the Desktop equivalent of Claude Code's file access — but limited to folders you explicitly authorise, and read-only in most configurations. It reduces the upload dance for files in those folders, but does not fully replace Claude Code's flexibility.

Installing a custom MCP server (for Zotero)

If the Zotero integration is not available in the built-in store, you can add it manually:

Settings → Developer → Add MCP Server → provide the server configuration

The Zotero MCP server configuration looks like:

{
  "command": "npx",
  "args": ["-y", "zotero-mcp"],
  "env": {
    "ZOTERO_API_KEY": "your-api-key-here",
    "ZOTERO_USER_ID": "your-user-id-here"
  }
}

(Your Zotero API key is found at zotero.org/settings/keys)

Practical workflow tips for Desktop

Use separate Projects for separate research threads. Do not put everything in one Project. A Project for each article, each grant, each reading list.

Put your current draft in the Project. Upload the latest version of your draft. When you start a conversation about the paper, Claude already has it.

Set Project instructions in the voice of a collaborator. Instead of "you are a helpful assistant", try: "You are a co-author on this article. You know the argument and the sources. Help me write and revise." This changes the register of responses.

Use the conversation history for your own reference. Desktop saves your conversations. They are searchable. Treat them as a log of your thinking — you can return to a conversation from three weeks ago.

Combine global and Project instructions deliberately. Global instructions: your identity, your general preferences, your citation style. Project instructions: the specific project context, the argument, the sources, the current task.

The limits of Desktop (and when to switch to Code)

Even with Projects and integrations, Desktop has structural limits:

These limits are not bugs — they are the result of Desktop being a web application rather than a local tool. When these limits become friction in your actual work, that is the moment to try Claude Code.

Related