Inciteful MCP: the citation graph inside your AI assistant
Connect Claude, ChatGPT, Cursor or any MCP client to Inciteful. It snowballs from seed papers, finds the important and review papers around them, and traces the paths between two papers. It’s free, needs no login, and runs on open OpenAlex data.
Add this URL to your assistant as a remote MCP server:
https://incitefulmed.com/academic/mcp Claude
- In Claude, open Customize → Connectors, click + Add, then Add custom connector.
- Name it “Inciteful”, paste
https://incitefulmed.com/as the server URL, and continue.academic/ mcp - Under Authentication, choose No sign in, then click Add.
- In a chat, click +, open Connectors and turn Inciteful on.
- Ask: “Using Inciteful, find papers similar to 10.1038/nature14539.”
Free plans allow one custom connector. On Team and Enterprise plans, an owner adds it under Organization settings → Connectors, and members connect it from Customize → Connectors.
Steps checked on 2026-10-30 against Anthropic’s custom connector guide.
ChatGPT
- Open ChatGPT’s Plugins page (chatgpt.com/plugins), select the + button, then Add custom MCP server.
- Enter the name “Inciteful” and a short description. Under Connection, choose a public endpoint and paste
https://incitefulmed.com/.academic/ mcp - For authentication, pick the option without sign-in: Inciteful needs no account. Review the warning, select I understand and want to continue, then Create as a plugin.
- In a new chat, type @, pick Inciteful and ask your question.
Account and workspace policies decide whether you can add custom MCP servers. In a business workspace, an admin may need to allow it.
Steps checked on 2026-10-30 against OpenAI’s guide to connecting an MCP server.
Cursor
- Open
~/.cursor/mcp.jsonfor all your projects, or.cursor/mcp.jsonin one project. - Add the server:
{ "mcpServers": { "inciteful": { "url": "https://incitefulmed.com/academic/mcp" } } } - Save the file. Inciteful appears with your MCP servers under Customize in the sidebar, where you can turn it on or off.
Steps checked on 2026-10-30 against Cursor’s MCP docs.
Claude Code
- Run:
claude mcp add --transport http inciteful https://incitefulmed.com/academic/mcp - That adds it for the current project. Add
--scope userto use it in all your projects. - Start
claudeand ask, for example: “Using Inciteful, find the important papers around 10.1038/nature14539.”
Steps checked on 2026-10-30 against Claude Code’s MCP docs.
VS Code
- Open the Command Palette and run MCP: Add Server, then choose HTTP (HTTP or Server-Sent Events).
- Paste
https://incitefulmed.com/, enter the IDacademic/ mcp inciteful, and choose where to save it (Global makes it available in every workspace). - Or add it to
.vscode/mcp.jsonyourself:{ "servers": { "inciteful": { "type": "http", "url": "https://incitefulmed.com/academic/mcp" } } } - Ask your question in the Chat view.
Steps checked on 2026-10-30 against VS Code’s MCP server docs.
Other clients
- Add a remote MCP server with the URL
https://incitefulmed.com/. The transport is Streamable HTTP.academic/ mcp - Leave authentication empty: no API key, header or sign-in.
- Clients configured with JSON often take
{"mcpServers": {"inciteful": {"url": "…"}}}, but key names vary, so check your client’s docs.
Steps checked on 2026-10-30 against the MCP transport specification.
- Free
- No account
- OpenAlex metadata (CC0)
- Open-source server (MIT)
What you can ask it to do
Find similar papers, fast
Rank papers by citation-pattern similarity to one seed paper or several.
find_similar_papers“What else should I read like these three papers?”
Explore the citation network
Build the citation network two steps out from your seeds, then rank the important, similar, review and recent papers, plus authors, journals and institutions. Each seed’s references and citations are there too.
explore_citation_network“Which foundational papers am I missing?”
Connect two papers
Show the citation paths and bridge papers between two works, even across fields.
connect_papers“How does this physics paper connect to that ecology one?”
Ask the graph in SQL
Run your own SELECT query over the network, with the same engine as the web app’s SQL view.
query_citation_graph“Top authors since 2020 by PageRank in this network.”
Find and read papers
Search by topic, fetch up to 100 records with abstracts, and read open-access full text from PubMed Central.
search_literature · get_papers · read_paperExport citations
Export BibTeX or RIS (for Zotero, EndNote and others).
export_citations“Export the top 30 as BibTeX.”
It also includes a literature-review prompt, literature_review, that runs the iterative snowball with you: resolve your seeds, find similar papers, agree 3–5 seeds, explore the network, read the finalists’ abstracts, then write with a DOI for every paper.
See it work
I have three papers on deep learning: 10.1038/nature14539, 10.1038/nature16961 and 10.1162/neco.1997.9.8.1735. Using Inciteful, what important work around them am I missing?
→ explore_citation_network · 3 seeds · analysis: important
Ranked by PageRank inside the citation network around your three seeds, the most central papers you don't cite yet are listed below, each with its DOI. The provenance block at the end records the graph snapshot, its size and the method, for your methods section.
Illustrative example, not a recorded run
How it ranks, and how you can report it
Use it to report citation searching (PRISMA-S, TARCiS).
- Similar: Adamic/Adar bibliographic coupling. Shared niche references count more than shared popular ones.
- Co-citation: the Salton index, normalized so famous papers don’t dominate.
- Important: PageRank inside your local network, not global citation counts.
- Fast similarity: one step out from the seeds, for quick answers.
- Connector: shortest undirected citation paths between two papers.
Every citation-graph result names the method that ranked it and carries a provenance block. The fields are explained on the methods page.
"provenance": {
"retrieved_at": "2026-10-30T18:02:11Z",
"tool": "explore_citation_network",
"data_source": "OpenAlex metadata (CC0) via Inciteful's citation graph",
"graph_snapshot": "2026-10-15",
"seeds": [
{ "input": "10.1038/nature14539",
"openalex_id": "W2919115771" }
],
"graph": {
"papers": 18240,
"depth": 2,
"prune_threshold": 10000,
"highly_cited_not_expanded": 3
},
"method": "page_rank (local)",
"filters": { "title_all": ["learning"] },
"sql_executed": "SELECT ... LIMIT 25"
}
Abridged; real results carry every field. Paste it into the methods appendix of a PRISMA-S or TARCiS report.
Data, coverage and limits
- Graph snapshot: 2026-10-31. Snapshot 2026-10-31, updated nightly through 2026-10-31 with new papers that have a Crossref DOI. Other new papers are looked up live only when named.
- Title filters match paper titles only (not abstracts or full text), with Porter stemming. The tools say so, so your assistant won’t over-filter.
- Very highly cited papers aren’t expanded, which keeps graphs a usable size. When that happens, the result says how many.
- Licences: metadata is OpenAlex CC0; full text from PubMed Central carries each article’s own licence, which the result names.
- Retraction and preprint flags from OpenAlex are passed through. Metadata errors in open indexes can still occur.
- Fair-use limits: per-caller rate limits keep it free for everyone.
Privacy
- No account, no sign-in and no cookies.
- Our usage log keeps, for 90 days, one row per tool call: the tool name, your client’s name and version, how long it took, the outcome and the graph size. Your IP address is stored as a hash with a salt that changes every day. The usage log holds no queries, paper IDs or raw IP addresses.
- Our graph service’s own logs may contain the paper IDs and queries it processes. We keep them for up to 15 days, with no IP addresses.
- Data goes to Cloudflare (network), OpenAlex (search terms and IDs), NCBI (full text), Sentry (our graph service’s error and performance monitoring: request addresses that may include paper IDs, no IP addresses, kept up to 90 days) and your assistant’s vendor, under its own policy.
- Don’t put personal or health details in your prompts.
Works well with
Zotero, EndNote and other reference managers, through the BibTeX and RIS export; and the Inciteful web app, where results link to an interactive graph you can keep exploring.
FAQ
Is this the same Inciteful as inciteful.xyz?
Yes. Inciteful now lives at incitefulmed.com/academic. The web app, the graph and the methods are unchanged, and it’s still free.
Who builds it, and why is it free?
The team behind Inciteful and Inciteful Med builds it. Inciteful has been free since it started; the MCP server runs on the same citation graph service as the web app, and fair-use limits keep the cost manageable.
Why is it on a medical domain?
Inciteful Academic moved to incitefulmed.com/academic, next to Inciteful Med. The graph covers every field in OpenAlex, not just medicine.
Which assistants work?
Claude (web, desktop, mobile and Claude Code), ChatGPT where your plan allows custom MCP servers, Cursor, VS Code, and any client that supports remote MCP over Streamable HTTP.
My assistant didn’t use Inciteful.
Name it in your prompt, for example “Using Inciteful…”. Assistants often default to web search.
Why are some recent papers missing?
Snapshot 2026-10-31, updated nightly through 2026-10-31 with new papers that have a Crossref DOI. Other new papers are looked up live only when named.
Why did I hit a rate limit?
Every Claude user shares one rate-limit bucket, because Claude’s requests reach us from Anthropic’s servers; the same goes for ChatGPT and OpenAI’s servers. If you see a rate-limit error, wait a minute and retry.
What happens if the service is down?
If the server is down, so is the search. The MCP server runs as a single service with no standby, so your assistant will report a connection error until it’s back; try again a little later.
Is it open source?
The server is, under the MIT licence, on GitHub. It calls Inciteful’s hosted graph service to build citation graphs, and that service isn’t part of the repository.