4Science Community call: develop an Open-Source MCP Server for DSpace Are you scared of AI and AI implementation everywhere? Don’t worry, we are too, and we are doing our best to get the best out of this AI wave that is hitting us. However, not everything about AI has to be scary or bad; sometimes it can help you in your daily work and improve it. That is why we wanted to take a step toward AI because, as we like to say, “If you can’t beat them, join them.” That is why we have released dspace-mcp, a PoC open-source server that connects AI assistants directly to DSpace repositories. Now available on GitHub, the project is the first implementation of the Model Context Protocol (MCP) for the DSpace ecosystem, enabling you to search, manage, and submit content to digital repositories through natural-language conversation.
What problem does this address?
4Science Community call: develop an Open-Source MCP Server for DSpace: Worldwide, universities, libraries and research institutions make use of DSpace, one of the most widely adopted open-source tools for managing digital collections consisting of scholarly papers, theses, datasets and institutional records. Yet usually interaction with these repositories has had to be carried out by means of navigating their web interfaces or by writing custom scripts for the DSpace REST API.
By using dspace-mcp, that gap is eliminated since a researcher, a librarian, or a repository administrator can now request an AI assistant to find a paper, upload a document, or update the metadata records and the assistant will carry out the action directly within DSpace.
DSpace is becoming an enabled system in agentic workflow
How It Works
The project is built in TypeScript and expose tools reusing the REST API of your DSpace repository. It acts as an adapter layer translating the MCP protocols to the DSpace REST API. It can be easily extended with additional tools wrapping any of the DSpace API. No further code to plugin in your DSpace codebase, no further complexity to deal with during upgrade. Easily to connect to any existing repository.
The tool descriptions are given in natural language so that the AI model knows when and how to use them; for instance, if a user asks ‘Find all papers about climate change that were published after 2020′, the AI looks at the tool descriptions, chooses dspace_search, fills in the relevant parameters, and then displays the results in a format that is readable by humans.
One of the most practical design decisions in this project is its dual-runtime architecture:
- Local mode (stdio): The server runs as a standard Node.js process and communicates over standard input/output. This mode is used when connecting to desktop AI tools like Claude Desktop or the Cursor code editor. A developer points their AI client to the server with a one-line configuration.
- Cloud mode (HTTP via AWS Lambda): The same server can be packaged as a Docker container and deployed to AWS Lambda behind an API Gateway, making it accessible as a remote HTTP endpoint. This is ideal for institutions that want to offer AI-powered repository access as a centralized, always-available service without running any local software.
By adopting this dual approach, one codebase is used both by individual researchers working from their laptops and by large organizations that operate cloud infrastructure.
- Configuration (config.ts)
Setup is deliberately minimal. The server needs just one environment variable, DSPACE_BASE_URL, pointing to the target DSpace REST API endpoint (for example, https://sandbox.dspace.org/server). An optional PORT variable controls the HTTP server port for cloud deployments.
What Can You Actually Do With It?
Here are some real-world scenarios that dspace-mcp enables:
- A researcher opens Claude Desktop and asks: “Search our institutional repository for all datasets related to genomics submitted in the last year.” The AI returns a structured list of results with titles, authors, and links. Download the RAW data and start analyzing it without leaving its AI notebook
- A librarian says to their AI assistant: “Create a new item in the ‘Articles’ collection using the information contained in the email received by the Dean about its last article.” The AI assistant retrieves the email content, authenticates on DSpace, creates the item, populates the metadata fields, and uploads the file all in one conversation.
- A repository administrator asks: “Update the subject keywords on item [UUID / handle / Title] to include ‘machine learning’ and ‘neural networks’.” The AI uses JSON Patch operations to surgically update only the specified metadata fields without affecting the rest of the record.
- A department coordinator uses the submission workflow: “Start a new submission in the ‘Faculty Publications’ collection with this paper’s metadata.” The AI creates a workspace item that enters the institution’s normal review and approval pipeline.
Why It Matters
Since it was first introduced, the Model Context Protocol has experienced rapid growth; by the middle of 2026 there had been more than 10,000 MCP servers in use around the world, and its SDKs are downloaded more than 97 million times each month. Major AI companies, such as OpenAI and Google DeepMind, have adopted the protocol.
Using dspace-mcp, 4Science is bringing this surge in AI interoperability to the field of cultural heritage, research, and the academic sector. The project enables AI assistants to gain structured, authenticated, and secure access to DSpace repositories, thus allowing for more efficient research workflows, faster cataloging, smarter discovery, and ultimately wider access to the world’s scholarly output.
The code is available now at: https://github.com/4Science/dspace-mcp