PaperGraph MCP
An MCP server that splits math papers from arXiv, LaTeX and PDF into theorems, proofs and dependencies backed by the source
- 278 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Calls Crossref, OpenAlex and arXiv for metadata and open PDFs
- Writes a local SQLite database with notes, queues and imported papers
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/papergraph-mcpDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Calls Crossref, OpenAlex and arXiv for metadata and open PDFs
- Writes a local SQLite database with notes, queues and imported papers
This is third-party code. Review the repository files before installing.
What it does
PaperGraph MCP turns an arXiv paper, a local LaTeX project or a born-digital PDF into a theorem-centered workspace. The agent opens a paper map: structure, main-result candidates and the places where reading depends on outside work. It then picks a result and inspects its statement, proof, source slices and dependencies. Dependencies are traced direct and recursive, with shared prerequisites, cycles and unknowns. The server does not guess: an empty dependency list does not prove independence, and a reading order is not a complete graph. External references become import plans; candidates are searched in Crossref, OpenAlex and arXiv, and a download happens only after an explicit choice. Queues, sessions, checkpoints, notes and open questions live in a local SQLite workspace outside the repository. Results export to Markdown as a reading report for one paper or a reading plan across several related papers. The same functions work from the command line with no MCP client.
Who it is for. For mathematicians, graduate students and researchers who read hard papers with an AI agent and want every claim to lead back to a line of the source.
Good fit when
- You need the structure of a new math paper and a result to start from
- You must trace which lemmas a theorem relies on and where a proof cites outside work
- Reading spans several sessions and needs queues, checkpoints and notes
- A set of related papers needs one Markdown reading plan
- You need to find an open full text by DOI without anything being downloaded automatically
Not a fit when
- You need proof verification: the server does not verify proofs or infer hidden prerequisites
- The paper is a scan without a text layer: born-digital PDFs are supported
- You need paywalled texts: there is no paywall bypass and metadata is not full text
- You need a complete dependency map of the whole work: automatic PDF proof evidence covers only the first line with an explicit proof heading
Example request
Load arXiv paper 2301.00001, show the paper map and pick the main result; then walk through its proof and dependenciesMCP
- Transport
- stdio
- Authentication
- not required
Limitations
The server does not verify proofs, infer hidden prerequisites or establish semantic theorem equivalence. For long PDFs, automatic proof evidence covers only the first physical line with an explicit proof heading, and ownership may stay unknown. External references become plans to review and nothing downloads on its own. The workspace is local; back it up before schema upgrades, because older versions cannot open schema 9. Python 3.10 or newer and uv are required. Network requests go to Crossref, OpenAlex and arXiv.
How to disable. Remove the `papergraph` block from the MCP client config and restart the client. Workspaces live in SQLite databases you created outside the repository; delete them if needed.
FAQ
How do I connect PaperGraph MCP to Claude Code or Codex?
Add a stdio server to the MCP config with command `uvx` and args `--from papergraph-mcp==1.2.3 papergraph-mcp`, then restart the client. Recipes for Codex and Claude are in `client-configuration.md`; the author has not yet checked them on real clients.
Why does the server stay quiet after launch?
With no arguments it waits for an MCP stdio connection and prints nothing. To check the install use `papergraph-mcp doctor` and `papergraph-mcp --version`.
What is a Paper Map?
It is the first view of a paper: structure, main-result candidates and the places where reading depends on outside work. It exists so you pick a theorem before reading its proof.
How do I find a full text by DOI?
The `discover-doi DOI` command looks for candidates and downloads nothing. `add-doi-paper WORKSPACE DOI` imports the root paper when exactly one eligible public PDF exists; with several candidates you must choose.
Where are notes and reading queues stored?
In local SQLite workspaces. Keep the database outside your project repository; the `plan-starter-project` command previews what would be created and writes nothing.
Discussion
No questions or reviews yet.
Related
MCP serversA local GraphRAG MCP server: it builds a knowledge graph from documents and answers with source citations through Ollama.
pwc CLI
pwc-cli
Hugging Face's CLI, skill and MCP server: an agent searches papers, benchmarks and top models in Papers with Code with no account or key
Graphiti MCP
Graphiti MCP Server
Zep's Graphiti MCP server: long-term agent memory as a time-aware knowledge graph on FalkorDB or Neo4j
An MCP server and filesystem for markdown: an agent writes notes while people read the same files in a web UI with git history
A universal MCP CLI client from Apify: persistent sessions, OAuth, JSON output for scripts and shell access to any MCP server for agents
A local MCP server that keeps a staged journal of an agent's reasoning with persistent sessions, decisions and evidence
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| PaperGraph MCP (this card) | no | not required | 278 | |
| arxiv-mcp-server | no | not required | 3,276 | search and section-by-section reading of arXiv papers, with no theorem or dependency parsing |
| Paper Search MCP | no | key is optional | 3,076 | search and download of papers from arXiv, PubMed, bioRxiv and other sources |
README in short
The README starts with the payoff: reading math with evidence. It then gives a quick uvx launch, the MCP config, a first paper walkthrough with commands, the evidence boundaries and links to documentation with sample reports and reading plans.