Research Scan — Claude Code MCP server: find and verify papers
Pulls 5–10 recent papers that fit a research question and checks every DOI against the live record. Useful when a decision needs literature behind it.
claude mcp add research-scan -- uvx research-scanWhat Research Scan does
The tool splits the job in two. A model-free CLI queries OpenAlex, Semantic Scholar and arXiv, follows citation links, drops duplicates, tallies coverage against each criterion and confirms DOIs against the live record. Judgement stays with the agent: it drafts the queries, scores candidates from 0 to 3, writes extra queries where coverage looks thin and reranks the shortlist. The scoring rubrics ship as plain Markdown under skills/research-scan/references/.
Files carry the handoff. Reasoning and execution talk through the filesystem, so Claude, Codex, Cursor or a plain Python loop can drive the same chain over the same artifacts. Each run leaves a directory of validated JSON plus a Markdown brief for a human, with title, year, first author and DOI of every paper checked against the live record.
You need Python ≥ 3.11 and uv. OPENALEX_API_KEY is mandatory — doctor exits 3 without it. OPENALEX_MAILTO lifts the OpenAlex rate and puts you in the Crossref polite pool; S2_API_KEY removes the throttling on Semantic Scholar. research-scan configure writes them to ~/.config/research-scan/.env.
What it helps with
- build a recent reading list around a decision, not a topic
- filter out candidates whose DOIs fail live verification
- spot thin coverage and run a second round of queries
- get a ranked shortlist instead of a pile of links
- hand a colleague a brief where every paper was checked against current metadata
Example prompts for the agent
- Run
/research-scanonresearch/my-brief.mdwith--profile standardand keep 10 papers - Use
/research-scanto ask which interventions show the strongest recent evidence on long-term savings - Do an
orientscan and tell me where coverage is still thin
How to install Research Scan
Platform CLI
claude mcp add research-scan -- uvx research-scanuvx runs the command named after the package; if the README names another one, use uvx --from research-scan <command>
Platform CLI
codex mcp add research-scan -- uvx research-scanuvx runs the command named after the package; if the README names another one, use uvx --from research-scan <command>
Config file
{
"mcpServers": {
"research-scan": {
"command": "uvx",
"args": [
"research-scan"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from research-scan <command>
Config file
{
"servers": {
"research-scan": {
"type": "stdio",
"command": "uvx",
"args": [
"research-scan"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from research-scan <command>
Platform CLI
gemini mcp add research-scan uvx research-scanuvx runs the command named after the package; if the README names another one, use uvx --from research-scan <command>
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about Research Scan
How do I install Research Scan in Claude Code?
Run claude mcp add research-scan -- uvx research-scan. Python ≥ 3.11 and uv must be present first.
What does Research Scan do?
It retrieves 5–10 recent papers for a question or project, expands them through citation graphs and verifies each DOI against the live record. The hosting agent plans queries and ranks the shortlist.
What does Research Scan need to run?
OPENALEX_API_KEY is required. OPENALEX_MAILTO raises the OpenAlex rate, S2_API_KEY removes Semantic Scholar throttling and NCBI_API_KEY speeds up PubMed. research-scan configure stores them.
