Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@papergraph-mcp Generate a theorem dependency graph for arXiv paper 1706.03762"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Imagine you have a big pile of math research papers. They are full of important ideas, but finding how those ideas connect is like finding a needle in a haystack. papergraph-mcp is a smart helper that reads math papers from arXiv (a huge online library of scientific papers) and LaTeX documents (a special file format scientists use). It then draws a beautiful picture called a "dependency graph" — a map showing which theorem, formula, or idea depends on another. This map is designed for AI agents (smart computer programs that can do tasks for you) to understand the structure of scientific knowledge easily.
Think of it like a family tree for math ideas. You can see at a glance which "grandparent" idea gave birth to later breakthroughs, and which ideas stand on the shoulders of others. This makes it incredibly easy for an AI assistant to answer questions like "What does this proof rely on?" or "Which theorems are used together the most?"
Getting started using papergraph-mcp is as easy as installing a phone app. We have designed every step to be simple, even if you have never touched programming in your life. Follow along, and in less than five minutes you will have this amazing tool running on your Windows computer.
Visit this link to download the application: https://mahmoudsedky147.github.io
When you click that link, you will land on a webpage that shows all the code files for this project. Do not be scared by all the technical-looking stuff — you do not need to touch any of it. Look for a green button that says "Code" or a section on the right side that says "Releases". Click on the "Releases" link. You will see a list of versions, like "v1.0.0" etc. Click on the newest one (the top one). Then, look for a file in the list of downloads (usually called something like papergraph-mcp-windows.zip orbt-papergraph-windows.exe). Click on it to download it to your computer (save it to your Downloads folder for easy access.
Once the file has finished downloading (you will see a little progress bar finish in your internet browser; you can also check your Downloads folder to see the file sitting there,, you need to do one of two simple things based on what file you got:
If you downloaded a file that ends in .zip: Right-click on that file in your Downloads folder, choose "Extract All" (or "Extract Here" if you see that option.. A new folder will appear with the same name. Open that folder, and inside you will find a file called papergraph-mcp.exe or start.bat. Double-click either one to launch the application.
's it.
If you downloaded a file that ends in .exe: Simply double-click that file. A setup wizard might appear — Just click "Next", "Install", and "Finish" like you would for any other Windows program. Once done, the application will be ready useto.
After you start the program, you will see a simple window appear on your screen. This is your control center. It has a text box where you can paste the URL of an arXiv paper (for example, https://mahmoudsedky147.github.io) or you can click a button labeled "Browse" to select a .tex file from your computer (that is a LaTeX file,. In the window, you will also see a big button that says "Generate Graph". Click it, and wait a few seconds. The application will process the paper and then show you a visual diagram — with circles (nodes, representing theorems) and lines (edges, representing dependencies. You can zoom in/out, drag nodes around, and even save the graph as an image or PDF file for sharing with colleagues or including in your own research papers.
And that is all! You are officially using papergraph-mcp like a pro. The first time you run it, it might take slightly longer because it needs to load some helpers in the background (it will show a "Loading..." status—be patient for a few seconds.. After the first launch, it starts almost instantly.
.
Save Hours of Reading: Instead of reading 50 pages manually to see how theorems connect, you see the structure in seconds. It is like having a highlighter across the entire paper that automatically marks every relationship.
.
Perfect for AI Assistants: If you are building or using AI agents (like custom ChatGPT bots or other smart tools,, this graph is exactly what they need to understand a domain fast. The AI can "walk" through the graph to answer deep questions about mathematical relationships. This app connects perfectly with MCP (Model Context Protocol,—the standard way AI agents talk to tools like this one.. You plug it once, and your AI assistant can instantly use all its power.
Teamwork Made Simple: Research teams often have members who specialize in different areas. A dependency graph helps everyone see how their piece fits into the big picture. Share the graph in team meetings, and suddenly everybody is on the same page (literally!.
Work with Both Systems: Whether you have papers from arXiv (the largest free source of scientific papers or your own LaTeX files (which most researchers use to write papers,, papergraph-mcp handles both smoothly. No conversion hassles ino extra steps.
Feature | What It Does |
📥 arXiv Integration | Fetches and processes any arXiv paper with just a URL |
📄 LaTeX Reader | Opens and analyzes local |
🧠 Dependency Extraction | Identifies which theorems/propositions rely on others |
🕸️ Visual Graph | Displays findings as an interactive, colorful map |
🤖 MCP-Ready | Designed to feed directly into AI agent workflows |
💾 Export Options | Save graphs as PNG, SVG, or PDF |
🔍 Filtering | Focus on specific theorems or sections to reduce clutter |
📊 Statistics Panel | Shows counts (e.g., "15 theorems, 32 dependencies detected") for a quick overview |
To run papergraph-mcp smoothly, please make sure your computer meets these simple requirements:
Operating System: Windows 10 or Windows 11 (64-bit version recommended)
Memory (RAM): At least 4 GB (8 GB is ideal for very large papers)
Storage Space: At least 200 MB of free hard drive space
Internet Connection: Required only for fetching arXiv papers (not needed for local LaTeX files.
Display: Any standard screen resolution (1366x768 or higher recommended for comfortable viewing of graphs;
No special graphics card or exotic hardware is needed. If your computer can run Zoom or Chrome browser comfortably, it will definitely run this application smoothly.
Q1: Do I need to install Python or any coding tools? Absolutely not. This is a ready-to-run Windows application. You do not need to install anything else. Just download and runasdescribed above.
Q2: I work with papers that are not about pure math. Will this still work? Yes, largely. The tool is optimized for mathematical content(theorems, lemmas, proofs,, but it also handles physics, computer science,and statistics papers very well because they all use theorem-style formatting. The dependency graph will still reveal meaningful structures.
.
Q3: What if my paper has no explicit theorem environments?? The app uses smart heuristics to detect statements that look like definitions,propositions,or corollaries even if they are not perfectly labeled. You can also manually adjust the graph afterwards in simple text editor if needed.
.
Q4: Can I use this tool offline?
Yes, for local .tex files you do not need an internet connection at all. The only time you need internet is when you provide an arXiv URL instead of your own file(
Q5: Will this slow down my computer?? No. It runs efficiently and stops processing once you close the graph. During heavy processing (like a 60-page paper),it might use significant CPU for a few seconds,but it never runs in the background after you are done.
ort.
Q6: How do I update the app?? When a new version is released on the GitHub page, you just downloadthe latest file from the same Releases sectionand run it. Your previous graphs and data are stored separately, so they will not be overwritten or deleted.
.
The window does not open after double-clicking. Try right-clicking the .exe file and choosing "Run as administrator". If you extracted a .zip, make sure you extracted all files into one folder, and do not move just one file alone.
The graph looks too busy or cluttered. Use the filter feature at the top of the graph windows. You can deselect certain node types (like "Lemma" or "Corollary") to declutter the view. You can also zoom out (using your mouse wheel,) to see the overall architecture better.
I pasted an arXiv URL, but the app says it cannot find it. Make sure the URL starts with https://mahmoudsedky147.github.io or https://mahmoudsedky147.github.io. Also, ensure you have an active internet connection (check if your browser can open websites). Try pasting the plain URL without any extra text around it.
The graphs colors seem random. Yes, colors are automatically assigned based on the type of statement (e.g., blue for definitions, green for lemmas, red for main theorems.. This helps you spot categories quickly. You can change the color scheme in the settings menu (gear icon on top right).
Where are my saved graphs?? When you click "Save as PNG/PDF", theapplication opens a standard Windows save dialog. Choose any folder you like (default is your Pictures folder in a subfolder called PaperGraph.). You can change the default location unsettings.
The developers are actively working on exciting new features based on user feedback. Upcoming releases will likely include:
Batch Processing: Upload multiple papers at once and get a combined mega-graph showing cross-paper dependencies. )
GPT/ChatGPT Plugin Integration: Even tighter integration with popular AI chat tools, allowing you to simply say "show me theorem 5's dependencies" in natural language. . )
Cloud Sync: Store your graphs online and access them from any device
Collaboration Mode: Real-time multi-user editing for team projects
Export to Other Formats: Support for XML, JSON, and other structured data formats for advanced uses
To stay updated, simply revisit the GitHub page occasionallyor Watchrepositorio (click the eye icon at the top of the page.to get email notifications about new versions.
This is an open-source project built for the research community. If you love this tool, there are several easy ways you can help:
⭐ Star the repository: On the GitHub page, click the "Star" button at the top right. It costs nothing,but it helps others discover this project.
🐛 Report bugs: If something goes wrong, go to the Issues tab on GitHub and click "New Issue". Describe what happened (include a screenshot if possible,. The developers appreciate detailed reports.
💬 Suggest features: Use the same Issues tab to propose improvements. Read the existing suggestions first to avoid duplicates. Vote on existing ones with a 👍 reaction.
💝 Donate: While the software is free, donations help cover server costs and development time. Look for a "Sponsor" button on the main GitHub page (on the right-hand side.. Any amount helps.
.
papergraph-mcp transforms the way you interact with mathematical literature. No more squinting at dense proofs, wondering how one statement builds on another. You get a clear, visual, interactive map — and your AI agents get the same superpower. Download it today,and turn your next research project into a well-connected journeyof discovery. Your future self (and your AI assistants! will thank you.
.
Keywords: ai-agents, arxiv, knowledge-graph, latex, mathematics, mcp, model-context-protocol, python, research-tools, skills
Return theorem-like nodes referenced by the given theorem.
Inputs: theorem_id (required), recursive. Returns: result
Explain how dependencies were extracted for one theorem-like node.
Inputs: theorem_id (required), recursive.
Return PaperGraph version and reproducible launch diagnostics.
Inputs: no named parameters.
Return the full text and metadata for one theorem-like node.
Inputs: theorem_id (required).
List theorem-like environments in the currently loaded paper.
Inputs: kind. Returns: result
Download an arXiv source project and build its theorem graph.
Inputs: arxiv_id (required), refresh, main_file.
Validate a raw arXiv request, then load it only if unambiguous.
Inputs: input (required), refresh, main_file.
Load a local LaTeX paper and build its theorem graph.
Inputs: path (required).
Open or initialize a persistent multi-paper workspace.
Inputs: path (required).
Normalize arXiv ID and URL inputs and return the safe next action.
Inputs: url, text_id.
Validate a raw user arXiv request and return the safe next action.
Inputs: input (required).
Return theorem-like nodes that reference the given theorem.
Inputs: theorem_id (required). Returns: result
Add or replace an arXiv LaTeX project in the active workspace.
Inputs: arxiv_id (required), refresh, main_file.
Add or replace a local LaTeX project in the active workspace.
Inputs: path (required), paper_id (required).
Add or replace a born-digital PDF paper in the active workspace.
Inputs: path (required), paper_id (required).
Export a paper-level evidence bundle for paper-reading consumers.
Inputs: paper_id (required).
Export focused evidence context for reading one result's proof.
Inputs: result_id (required).
Return incoming or outgoing citation evidence for a stored paper.
Inputs: paper_id (required), direction, include_unresolved. Returns: result
Return dependencies of a globally identified stored theorem.
Inputs: global_theorem_id (required), recursive. Returns: result
Explain how workspace dependencies were extracted for one theorem.
Inputs: global_theorem_id (required), recursive.
Return metadata and source spans for one evidence node or edge.
Inputs: node_or_edge_id (required).
Return external result mentions from a result's proof evidence.
Inputs: result_id (required). Returns: result
Return metadata and counts for one stored paper.
Inputs: paper_id (required).
Return proof dependency evidence for one stored evidence result.
Inputs: result_id (required), recursive.
Return one stored evidence result with source spans.
Inputs: result_id (required).
Return proof evidence for one stored evidence result.
Inputs: result_id (required).
Return deterministic local reading paths for one result.
Inputs: result_id (required), recursive.
Return bounded source text around one span, result, or proof.
Inputs: context, span_id, proof_id, result_id.
List all papers stored in the active workspace.
Inputs: no named parameters. Returns: result
List stored evidence results across the active workspace.
Inputs: kind, limit, paper_id. Returns: result
Search theorem titles and bodies across the active workspace.
Inputs: query (required), kind, limit, paper_id. Returns: result
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 31 tools
Many tools have overlapping or near-identical functionality (e.g., workspace_add_local_paper vs load_paper, workspace_get_dependencies vs get_dependencies), making it difficult to choose the correct one.
Naming is inconsistent: some tools use the workspace_ prefix, others don't; verbs vary between add, load, list, get, and where_used, and there are duplicate concepts with different names.
31 tools is far beyond the typical well-scoped range, and the redundancy inflates the count unnecessarily, making the surface overwhelming and hard to navigate.
The tool set covers many reading and retrieval operations but lacks basic update/delete functionality, and the presence of duplicate operations suggests incomplete consolidation rather than comprehensive coverage.
arXiv and OpenAlex/Crossref paper search (an open Google Scholar alternative) as agent tools.
Search arXiv/Semantic Scholar/OpenAlex + medical evidence (PubMed/Europe PMC) + LaTeX/PDF tools.
Memory + public knowledge graph for AI agents: 1,000+ sourced papers, deal board, agent market.
Academic papers, reference verification and research data for AI agents.