Academic research involves frequent research findings. Finding papers, codes, related models, and datasets. This usually means switching platforms like Arxiv, Github, Hugging Face and more, and stitching connections together manually.
The Model Context Protocol (MCP) is a standard that allows agent models to communicate with external tools and data sources. For research discovery, this means that AI can use research tools through natural language requirements to automate platform switching and cross-reference.
Research Discoveries: Three Layers of Abstraction
Just like software development, research discoveries can be assembled from the perspective of layers of abstraction.
1. Manual study
At the lowest level of abstraction, researchers search manually and cross-reference manually.
1. Find an Arxiv2 paper. Search Github for Implementation 3. Check the face of the hug for Model/Dataset 4. Cross-reference authors and quotations 5. Manually organize your survey results
This manual approach can be inefficient when tracking multiple research threads or conducting systematic literature reviews. The repetitive nature of platform-wide search, metadata extraction, and cross-reference information naturally leads to automation through scripts.
2. Scripted Tools
Python scripts automate research discovery by processing web requests, analyzing responses, and organizing results.
def granked_research_info(Paper_url): Paper_data = scrape_arxiv (paper_url) github_repos = search_github (paper_data (‘title’)) hf_models = search_huggingface(paper_data(“author”)))
return Consolidate_Results(Paper_data, github_repos, hf_models) results = ghater_research_info(“https://arxiv.org/abs/2103.00020”))
Research trackers show systematic research findings built from these types of scripts.
Scripts are faster than manual investigations, but often due to changes in APIs, rate limits, or analysis errors, data cannot be collected automatically. Without human monitoring, the script could miss relevant results or return incomplete information.
3. MCP Integration
MCP makes these same Python tools accessible to AI systems through natural language.
#Example of research order
Find the latest trans architecture papers that have been published in the last six months.
– You need the available implementation code
– Focus on papers containing preprocessed models
– Include performance benchmarks if available
AI will assemble multiple tools and provide information gaps and reasons for the outcome.
user: “This paper finds all the relevant information (code, model, etc.): https://huggingface.co/papers/2010.11929”
AI:
This can be seen as an additional layer of abstraction on scripts where “programming languages” are natural languages. This follows the analogy of Software 3.0, where the direction of natural language research is software implementation.
This comes with the same warnings as the script.
Faster than manual investigations, but error-prone, without the quality of human guidance, depends on understanding implementation.
Setup and use
Quick Setup
The easiest way to add a research tracker MCP is to hug your Face MCP settings.
Go to huggingface.co/settings/mcp and search for “Research-Tracker-MCP” in available tools, add it to the tool, and follow the provided setup instructions for a particular client (Claude Desktop, Cursor, Claude Code, VS Code, etc.).
This workflow utilizes a hugging face server. This is the standard way to use face spaces to hug as an MCP tool. The (Settings) page is automatically generated and always provides the latest, client-specific configuration.
learn more
Let’s get started:
Building yourself:
community:
Ready to automate your research discovery? Try out the Research Tracker MCP or create your own research tools using the resources listed above.
