CiteProof documentation
CiteProof turns a research question into a structured literature review. These docs explain what the product does, how the research workflow runs, and how to connect the Claude Science audit layer.
What CiteProof does
CiteProof searches academic sources, screens the records it finds, extracts evidence into a table, checks DOIs and citations, and writes a structured paper. Research Mode keeps a human in the loop: you approve the scope, the literature and framework, and the hypotheses before the paper is assembled.
There are two ways to use CiteProof:
In the app
Research assistant
Run DeepSearch for a fast cited report, or Research Mode for a full paper with PRISMA 2020 tracking, an evidence table and three review gates.
Next to your agent
Claude Science + CiteProof MCP
Keep Claude Science as the researcher and use CiteProof MCP as the audit layer: protocol, PRISMA counts, screening decisions, evidence quotes and one export.
Choose your workflow
If you want CiteProof to search and draft, open the research app. Start with a question, choose DeepSearch or Research Mode, and approve each review gate when it appears.
If you already work in Claude Science and want the review decisions recorded and checked, use the MCP server. The agent still searches, reads and writes; the server stores its decisions, validates their shape, checks the quotes it is given and computes the review counts.
- Set up CiteProof for Claude Science: the Claude Desktop extension, a local npx install and the Claude Science local command.
- Read the MCP reference: all 24 tools, their arguments, limits, results and troubleshooting.
Your first run
- Create an account. Sign in, then open the app. The Free plan is enough to try a run; paid checkout is not open yet.
- Write one review question. Keep it answerable from published literature and narrow enough to screen.
- Choose a mode. DeepSearch returns a fast cited report. Research Mode runs the longer pipeline and pauses at three human checkpoints.
- Approve the review gates. Check the scope and keywords, then the literature and framework, then the hypotheses. Nothing downstream is committed without your approval.
- Check the output. Read the cited claims against their sources, review the evidence table, and export the report only after the counts add up.
Model output can be wrong. CiteProof checks structure, quotes and citation matches where it can; it does not prove that a claim is scientifically true.
Where to go next
- Research workflow: the pipeline, the sources, the review gates and what stays manual.
- Plans and limits: plan prices, allowances, resets and bring-your-own-key behaviour.
- Claude Science setup: install CiteProof MCP in the client you use.
- MCP tool reference: exact arguments, limits and result shapes for every tool.
- Troubleshooting: common setup and run problems, and how to report a reproducible bug.
Keeping these docs accurate
The MCP pages are written from the server's own specification and tool code. If the implementation and a docs page disagree, the implementation is authoritative: MCP/docs/SPEC.md and MCP/docs/CLAUDE_SCIENCE.md.
Product prices and allowances shown in the app come from the live plan catalog. The pricing page is the public summary and always takes precedence over a number copied into a guide.