Zotero is a widely used reference manager that stores your PDFs, annotations, and notes on papers, and senior academics often have libraries containing thousands of PDFs. In this tutorial, we’re going to connect Claude code with the Zotero library so that you can search literature, extract annotations, move your papers, and do all of this locally on the full-text PDFs of your papers, which you have collected over the years.
Learn which tools can speed up your research: This free 21-day email series introduces you to a range of academic tools and how they work together to speed up the time needed for a literature review. Most of these tools have been around for less than a few years, some use AI. Knowing them gives you an almost unfair productivity advantage for your research.
Cloud CoWork is a very powerful AI tool that sits on your desktop installation and allows you to work with your local files, Obsidian nodes, and, in this tutorial, your Zotero installation. If you’re unfamiliar with cloud, then follow this tutorial to learn the basics and the second part to learn about skills.
To connect Zotero to Claude co-work, we are going to follow these six steps:
- Install the Claude Desktop app
- Obtain an API key and User ID from Zotero
- Enable local connections in Zotero
- Install the Zotero MCP Connector and add the API key
- Launch Claude Code and check for the installation
- Use Claude to ask questions about your Zotero library.
Step 1: Install Claude Desktop
Claude has a desktop app which can be installed on any Windows, Mac, or Linux machine. You can download it under this link and install it according to the instructions.
Step 2: Obtain an API Key & UserID from Zotero
Log in to your Zotero account. Click on Settings (1) and select the Security tab, scroll down to “Applications” and click the “Create new Private Key ” (2) button.
Note the User ID that is displayed right under the “Create new private key” button (3) you will need it later in the installation.
In the new key dialogue, you can define permissions associated with this key (4). For instance, you might decide to give Claude read-only access to your library, so it does not accidentally break anything. Here is how it looks:
The key you generate is like a password and allows anyone with access to it to read and possibly modify your Zotero library. Treat it confidentially and don’t share it online.
Note down the key and userID before you proceed.
Step 3: Enable local connections in Zotero
Next, you need to go to Zotero (ensure you have Zotero 8 or later), click on Settings, select the Advanced Tab (1) and put a check mark next to “Allow application on this computer to communicate with Zotero” (2).
Step 4: Install the Zotero MCP Connector and add the API key
An MCP server acts as an interpreter between Zotero and the large language model, and the MCP connector is a small configuration file that contains the necessary settings. There are multiple MCP connectors for Zotero, but the most popular one is “zotero-mcp“, it also offers most functionality and allows us to link to a local Zotero installation (which contains all PDFs). So let’s run through the installation process:
Step 4.1 Download the Installer (Mac OS):
Installing an MCP connector is a technical process. Luckily, for Zotero MCP, the community has created convenient installation scripts and even a graphical installer for macOS. Go to this repository. If you’re on a Mac, click on the latest release under Releases (1) and scroll down to download the installer dmg file (2).
After this is done, after the downloading of the DMG file, just double-click it and launch the installer.
If you are on Windows, simply scroll down and follow the instructions. You need to download a script first somewhere on your computer (1) and then right-click it, selecting “Run with PowerShell” (2):
Step 4.2 Provide the API Key and UserID
The installation process will ask you for an API key and a user ID (1) from Step 2, add them here. There is not much that you need to configure during this process if choosing to use the default configuration, but depending on the size of your library, this can take quite some time.
The installation process will build a search database that allows you to search the full text of the PDFs. This is typically done with a very lightweight AI model that runs locally during installation. The quality of this model is sufficient, but if you notice the search is not working correctly, you could rerun this process by providing an API key to your Google or ChatGPT model. This might cost a few dollars, but it will generate better results. You will also need only one setup, and whenever you introduce new PDFs to your Zotero library. In that case, during the installation process, click on the advanced configuration option and provide the necessary API keys (this will likely be unnecessary for most).
Step 5: Check the installation in Claude
If Claude has been running during the installation, restart it. Navigate to the Customise button (1) in the sidebar and select connectors (2). The resulting Connector colleciton should display “Zotero (Local Dev)” (3):
At this point, everything is correctly installed, and you are ready to use the connector. Note that this list of tools displayed under the connector (3) describes exactly what you can do with this connector. For instance, “zotero-search-notes” allows you to search your Zotero library for notes based on your query. Claude will pick some of these skills and use them to complete the task you give it.
Step 6: Using The Zotero Connector in Claude
There’s nothing else you need to do in order to activate your connector. Just ask Claude any question regarding Zotero, and it will start using this connector to access your library. For example, you can ask which paper was added last:
Notice how Claude will find the right tool to execute your request. These are exactly the tools you have seen in the connector window in step 5.
What can you do with the Zotero Claude Connector?
By itself, the Claude Connector allows Claude to do the following things:
Search & Discovery
- Search by topic, keyword, title, or author
- Semantic search by meaning (e.g. “papers about sleep and memory” — finds relevant results even without exact title matches)
- Search across multiple criteria simultaneously (author + date + topic)
- Find the most recently added papers
- Find papers related to a paragraph you’re writing (paste text, get relevant references)
- Browse papers within specific collections
- Search across personal and shared/group libraries
Reading & Extracting Content
- Read full text of PDFs (not just metadata)
- Retrieve abstracts and paper details
- Show existing highlights and annotations
- Extract table of contents/outline from a PDF
Writing & Annotating
- Create highlight annotations on PDFs (with colour control)
- Create and attach notes to papers (e.g. summary, key findings, limitations)
- Update metadata (title, authors, date, abstract, tags)
Organising
- Create new collections (folders)
- Add or remove papers from collections
- Add, remove, or batch-apply tags
- Find duplicate papers
- Merge duplicates (with preview before committing)
Adding References
- Add papers by DOI (auto-fetches metadata and open-access PDF)
- Add papers by URL (e.g. arXiv links)
- Import local PDF files from your computer
- Chain multiple actions in one request (e.g. find seminal papers → import them → create a collection → organise them)
But the biggest benefit of connecting Zotero to Claude is that you now have access to Claude’s powerful sketch features, like the various skills that you download or program yourself, and if using Claude CoWork, you can combine it with your notes stored locally in Obsidian (see for instance the conceptual lit review workflow), papers you are working on, etc.
What is an MCP Server?
Software like Zotero communicates through APIs, a rigid, technical interface that accepts precise, structured commands. Large language models, by contrast, work in natural language. An MCP server bridges this gap. It receives your plain-language request, breaks it down into the specific API calls the software understands, executes those calls, and then passes the results back to the model, which translates them into a response you can actually read and use.
Summary
In this tutorial, we used Zotero MCP, which created a bridge between our local Zotero installation and local cloud installation. You’ll have to redo it on every machine that you want to use this process with, but the advantage is that you can work directly on the PDFs stored on this local installation, even add and remove annotations.
