MCP Server · priorwork.fyi

Quick start

One command. No key needed to start.

priorwork-mcp is a small Model Context Protocol server. It adds semantic search over papers from the major ML, NLP and computer-vision venues (NeurIPS, CVPR, ICML, ICLR, ACL, EMNLP, ICCV, AISTATS and more) to your agent's tools, so there's no browser tab and no copy-pasting abstracts. It uses a shared public key until you set your own.

# Claude Code
claude mcp add priorwork -- uvx priorwork-mcp

For Cursor or any other MCP client, add this to your MCP configuration:

{
  "mcpServers": {
    "priorwork": {
      "command": "uvx",
      "args": ["priorwork-mcp"]
    }
  }
}

It runs through uvx, so the only thing you need is uv. Nothing to install or keep updated yourself.

Tools

What your agent can call.

Tool	What it does
`search_papers`	Semantic search from a text query (a descriptive sentence or two works a little better than keywords). Returns ranked papers with authors, venue, year, similarity score, download link and full abstract. Pass `summarize=true` for a structured LLM summary of each paper: problem, core insight, method, key results, positioning.
`search_by_pdf_url`	Give it a public PDF link (an arXiv PDF, say) and get similar papers from the index.
`search_research_topics`	A different corpus: open research problems mined from the limitations authors state about their own work, gaps several independent papers ran into that nobody has closed. Describe what you want to work on in prose; keywords match nothing here, because each problem is stored as a synthetic abstract. Returns the problem, why it matters, what changes if it works, the risk, and concrete ways in.
`list_indexes`	Lists the paper indexes and their sizes.

The research topics are cached locally and refetched only when the server's artifact id changes, so repeated searches cost one small request rather than a 4.6 MB download each.

Configuration

Everything is optional.

Set these as environment variables: with -e KEY=value in claude mcp add, or in an "env" block in JSON configs.

Variable	Description
`PRIORWORK_API_KEY`	Your own key. Defaults to the shared public key, whose limit is shared by everyone on the default install.
`PRIORWORK_API_URL`	Override the API address. Defaults to `https://priorwork.fyi`.

Public key (the default) pw_live_65d65b6a1f42920fda777e306f0005c0b9141bba4327dff203be010c1f34be72

The shared key allows 500 requests a day in total, across everyone using the default install. If you're hitting that limit, get in touch for a key of your own and set it as PRIORWORK_API_KEY:

# Claude Code, with your own key
claude mcp add priorwork \
  -e PRIORWORK_API_KEY=your_key_here \
  -- uvx priorwork-mcp

The same key works with the REST API.

Example

Ask for prior work in plain words.

Find prior work on spatiotemporal forecasting with graph neural networks

1. Taming Local Effects in Graph-based Spatiotemporal Forecasting — Andrea Cini, et al. (2023 · NeurIPS)  [score: 0.802]
2. FourierGNN: Rethinking Multivariate Time Series Forecasting from a Pure Graph Perspective — ...

It's the check worth running before starting new work, or before you submit. A topic can gather over a thousand papers in a single year (see the Paper Timelapse), more than anyone can follow by hand.