Scholar Feed MCP Server

Verify your Scholar Feed API key is working. Returns connection status, subscription plan, key name, and today's API usage count.

Search Scholar Feed's 560k+ CS/AI/ML paper corpus by keyword. Returns papers with LLM-generated summaries, novelty scores, and structured extraction data (method, task, contribution type). Supports filtering by category, novelty, recency, method, task, dataset, contribution type, and whether papers have benchmark results.

Get full details for a single paper by arXiv ID. Returns title, authors, year, LLM summary, novelty score, links, and structured extraction data (method_name, contribution_type, task_category, datasets, baselines). Use fields='abstract' to include the abstract. Use get_paper_results for benchmark scores, or fetch_fulltext with sections='all' for the full paper content.

Find papers similar to a given paper. Uses precomputed bibliographic coupling + embedding similarity (updated daily).

Get the citation graph for a paper. 'citing' = outgoing references this paper cites; 'cited_by' = incoming citations from other papers.

Get today's trending CS/AI papers ranked by a composite score of recency, citation velocity, and institutional reputation. Papers from the last 7 days.

Extract paper content from an arXiv paper's LaTeX source. Two modes: 'results' (default) returns 800 chars of results/experiments + 3 table captions. 'all' returns full paper sections (abstract, introduction, related work, method, results, conclusion) at up to 3000 chars each + 5 table captions. ~62% of arXiv papers have LaTeX source. May take a few seconds.

Look up multiple papers at once by arXiv ID. Returns details for found papers and lists not-found IDs.

Get the GitHub repository summary for a paper — README content and file tree. Only works for papers with an associated code URL.

Export BibTeX entries for one or more arXiv papers. Returns formatted BibTeX text ready for use in LaTeX documents or reference managers.

Run a deep research session on a topic. Searches 512k+ CS/AI papers, synthesizes findings with an LLM into a structured report with clusters, gap analysis, and evidence chains. Takes 60-300 seconds depending on depth. Note: may take 60-300s. The 'quick' depth (~60s) is most reliable. Returns the full structured report as JSON.

Ask a follow-up question on a completed deep_research report. Finds new papers not seen in the original report and synthesizes a focused follow-up analysis. Requires the report_id from a previous deep_research call. Takes 20-60 seconds.

Get structured benchmark results for a paper. Returns quantitative results extracted from the paper: datasets evaluated, metrics, numeric scores, model comparisons, and baselines. Use this after get_paper to see how a paper performed on benchmarks.

Get the SOTA leaderboard for a dataset/benchmark (e.g. ImageNet, MMLU, GSM8K, SWE-bench). Returns top methods/models ranked by score. Only includes papers with absolute numeric results. Powered by 59k+ extracted benchmark results across 20k+ datasets.

Search for datasets/benchmarks by name. Returns matching benchmark names with paper counts and available metrics. Use this to find the exact benchmark name before calling get_leaderboard. Covers 20k+ datasets from 24k+ papers.

Search papers by method or technique name (e.g. 'LoRA', 'YOLO', 'DPO', 'attention'). Unlike keyword search, this searches the structured method_name field extracted from 78k+ papers. Returns papers that introduce or evaluate the method, with benchmark result counts.

Compare 2-10 models/methods side-by-side across shared benchmarks. Finds datasets where at least 2 of the specified models have been evaluated, enabling direct score comparison. Example: compare GPT-4, LLaMA-3, and Mistral across MMLU, GSM8K, etc.

Discover researchers by topic (semantic search) or name. For research topics like 'efficient LLM inference' or 'graph neural networks', uses embedding similarity to find relevant authors. For short name queries, uses fuzzy name matching. Returns h-index, paper counts, research topics, and rank scores.

Get detailed author profile by ID (from discover_authors results). Returns h-index, total citations, global rank, research topics, novelty scores, and their top 10 papers by rank score.

Get all papers by an author (paginated, sorted by rank score). Use discover_authors to find the author_id first. Returns the same paper fields as search_papers.

Get raw benchmark score data points over time for a dataset+metric. Returns individual (paper, date, score, value_string) entries ordered chronologically. No trend lines or interpretation — raw scatter data. Use search_benchmarks first to find the exact dataset and metric names.

Get score distribution statistics for a dataset+metric across all papers. Returns min, max, median, mean, p25, p75, stddev, and count. Use this to contextualize a paper's claims — e.g., 'For MMLU accuracy, the median is 72.5% across 45 papers, range 33%-95%.' No judgment or outlier flags — just raw statistics.

Get aggregated research landscape statistics for a topic. Uses semantic search to find relevant papers, then returns count-based aggregates: methods used (with paper counts), benchmarks evaluated (with paper counts), active authors, contribution type distribution, publication velocity by month, and novelty score distribution. All data is factual counts — no rankings or editorial labels.