verify_citation is the tool that matters here: an assistant that can confirm a reference exists, then walk outward through the citation graph, stops producing plausible-looking references nobody wrote. The public tools answer without a plan, so the checking half costs nothing to try.
PapersFlow's hosted research server at https://doxa.papersflow.ai/mcp, over Streamable HTTP. A public read-only tool surface covers search and citations; account and paid workflows sit behind a PapersFlow plan.
- Literature search and record retrieval — search, search_literature, fetch
- A citation checked against the record it claims — verify_citation
- Related work found from a paper you already have — find_related_papers
- The citation graph around a paper, its immediate neighbours, and an expanded view — get_citation_graph, get_paper_neighbors, expand_citation_graph
- Signed in: evidence summarised across sources — summarize_evidence
- On a plan: deep scans run, polled, snapshotted and read back, plus Python plots — run_deepscan, get_deepscan_status, get_deepscan_live_snapshot, get_deepscan_report, run_python_plot
A PapersFlow account. Connect over Streamable HTTP to https://doxa.papersflow.ai/mcp and complete OAuth in the browser, or send a personal access token as an Authorization bearer header. The mcp.public.read scope covers search and citation work; mcp.research.read and mcp.research.run open the plan-only tools. Codex, Claude Code and Gemini CLI are tested against production.
One command plus a key — claude mcp add --transport http papersflow-mcp https://doxa.papersflow.ai/mcp --header "Authorization: Bearer YOUR_TOKEN", then supply credentials