OpenReview MCP server

A Model Context Protocol (MCP) server that provides access to OpenReview data for research and analysis. This server allows you to search for users, fetch papers, and export research data from major ML conferences (ICML, ICLR, NeurIPS).

Features

• User search: Find OpenReview profiles by email address
• Paper retrieval: Fetch all papers by a specific author
• Conference papers: Get papers from specific venues (ICLR, NeurIPS, ICML) and years
• Keyword search: Search papers by keywords across multiple conferences
• JSON&PDF export: Export search results to PDF and JSON files for convenient reading or further analysis and coding assistant usage

Installation

1. Clone the repository

git clone https://github.com/yourusername/openreview-mcp-server.git
cd openreview-mcp-server

2. Create and activate virtual environment

python3 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install the package

pip install -e .

Configuration for Cursor

Step 1: Get your OpenReview credentials

You'll need your OpenReview account email and password.

Step 2: Find your Cursor MCP configuration file

Either run cmd+shift+P to open the Command Palette and find MCP settings that will lead you to the mcp.json, or look for:

Cursor: ~/.cursor/mcp.json

Step 3: Add the OpenReview MCP server

Open the MCP configuration file and add the openreview server to the mcpServers section:

{
  "mcpServers": {
    "openreview": {
      "command": "/ABSOLUTE/PATH/TO/openreview-mcp-server/venv/bin/python",
      "args": ["-m", "openreview_mcp_server"],
      "cwd": "/ABSOLUTE/PATH/TO/openreview-mcp-server",
      "env": {
        "OPENREVIEW_USERNAME": "your_email@domain.com",
        "OPENREVIEW_PASSWORD": "your_password",
        "OPENREVIEW_BASE_URL": "https://api2.openreview.net",
        "OPENREVIEW_DEFAULT_EXPORT_DIR": "./openreview_exports"
      }
    }
  }
}

Important:

• Replace /ABSOLUTE/PATH/TO/openreview-mcp-server with the actual path (e.g., /Users/yourname/workspace/openreview-mcp-server)
• Replace your_email@domain.com and your_password with your OpenReview credentials
• Use the full path to the venv Python interpreter (ending in /venv/bin/python)

Example configuration:

{
  "mcpServers": {
    "openreview": {
      "command": "/Users/john/workspace/openreview-mcp-server/venv/bin/python",
      "args": ["-m", "openreview_mcp_server"],
      "cwd": "/Users/john/workspace/openreview-mcp-server",
      "env": {
        "OPENREVIEW_USERNAME": "john@university.edu",
        "OPENREVIEW_PASSWORD": "mySecurePassword123",
        "OPENREVIEW_BASE_URL": "https://api2.openreview.net",
        "OPENREVIEW_DEFAULT_EXPORT_DIR": "./openreview_exports"
      }
    }
  }
}

Step 4: Restart Cursor

Completely quit and reopen Cursor for the MCP server to load.

Usage

Once configured and Cursor is restarted, you can use natural language to interact with the OpenReview MCP server:

Example queries:

Search for papers:

Search OpenReview for papers about "multimodal tokenization" from ICML 2025, ICLR 2025 and NeurIPS 2025

Get your own papers:

Get my papers from OpenReview using email researcher@university.edu

Export papers with PDFs:

Export papers about "multimodal tokenization" from ICLR 2024, download PDFs and extract text

Get conference papers:

Show me all papers from NeurIPS 2024

The server will automatically:

• Fetch papers from OpenReview
• Search across titles, abstracts, and authors
• Download and extract text from PDFs
• Export results to JSON for further analysis

Exported files are saved to ./openreview_exports/ by default (or your custom directory).

Example output

Available tools

search_user

Find a user profile by email address.

search_user(email="researcher@university.edu", include_publications=true)

get_user_papers

Fetch all papers published by a specific user.

Input schema:

Field	Type	Description	Required	Default	Allowed Values
email	string	Email address of the user whose papers to fetch	Yes	—	—
format	string	Format of the response: summary or detailed	No	summary	summary, detailed

get_user_papers(email="researcher@university.edu", format="detailed")

get_conference_papers

Get papers from a specific conference and year.

Input schema:

Field	Type	Description	Required	Default	Allowed Values
venue	string	Conference venue (e.g., "ICLR.cc", "NeurIPS.cc", "ICML.cc")	Yes	—	ICLR.cc, NeurIPS.cc, ICML.cc
year	string	Conference year (e.g., "2024", "2025")	Yes	—	Four-digit year (e.g., 2024)
limit	integer	Maximum number of papers to return	No	50	1–1000
format	string	Format of the response: summary or detailed	No	summary	summary, detailed

get_conference_papers(venue="ICLR.cc", year="2024", limit=50)

search_papers

Search for papers by keywords across multiple conferences.

Search modes:

• any: returns papers that match at least one of the keywords in the specified fields. If any keyword is found, the paper is included.
• all: returns papers that match all of the keywords in the specified fields. Only papers containing every keyword are included.
• exact: returns papers that contain the exact phrase (all keywords together, in order) in the specified fields.

Input schema:

Field	Type	Description	Required	Default	Allowed Values
query	string	Keywords or phrase to search for (e.g., "time series token merging", "neural networks")	Yes	—	—
venues	array	List of conference venues and years to search in. Each item: • venue: string • year: string	Yes	—	—
search_fields	array	Fields to search in. Options: "title", "abstract", "authors"	No	["title", "abstract"]	"title", "abstract", "authors"
match_mode	string	How keywords are matched: • "any": match any keyword • "all": match all keywords • "exact": match exact phrase	No	"all"	"any", "all", "exact"
limit	integer	Maximum number of results to return	No	20	1–100
min_score	number	Minimum match score (between 0.0 and 1.0)	No	0.1	0.0–1.0

search_papers(
  query="time series token merging",
  match_mode="all",
  search_fields=["title", "abstract"],
  venues=[
    {"venue": "ICLR.cc", "year": "2024"},
    {"venue": "NeurIPS.cc", "year": "2024"}
  ],
  limit=20
)

export_papers

Export search results to JSON files for analysis.

Input schema:

Field	Type	Description	Required	Default	Allowed Values
query	string	Keywords to search for before export	Yes	—	—
venues	array	List of conference venues and years to export from. Each item: • venue: string • year: string	Yes	—	—
export_dir	string	Directory to export JSON files to	No	./openreview_exports	—
filename	string	Base filename for the export (without extension)	No	auto-generated	—
include_abstracts	boolean	Whether to include full abstracts in export	No	True	True, False
min_score	number	Minimum match score for search results (0.0 to 1.0)	No	0.2	0.0–1.0
max_papers	integer	Maximum number of papers to export and download	No	3	1–10
download_pdfs	boolean	Whether to download PDFs and extract full text content	No	True	True, False

export_papers(
  query="neural networks",
  venues=[
    {"venue": "ICLR.cc", "year": "2024"},
    {"venue": "ICML.cc", "year": "2024"}
  ],
  max_papers=1, 
  download_pdfs=true, 
  include_abstracts=true,
  export_dir="./research_exports"
)

Example workflow

1. Search for papers on a topic of interest:

search_papers(query="time series forecasting", match_mode="all", venues=[{"venue": "ICLR.cc", "year": "2024"}])

2. Export relevant papers to JSON:

export_papers(query="time series token merging", venues=[{"venue":"ICML.cc","year":"2025"}], max_papers=1, download_pdfs=true, include_abstracts=true)

3. Use the exported JSON files with Claude Code to implement methods inspired by the research.

Supported conferences

• ICLR (International Conference on Learning Representations)
• NeurIPS (Conference on Neural Information Processing Systems)
• ICML (International Conference on Machine Learning)

License

MIT License