1. ICML Paper Search MCP
ICML Paper Search MCP is an open-source paper search service that lets AI agents search 6,341 ICML 2026 papers.
The MCP provides six tools:
| Tool | Purpose |
|---|---|
open_search_session | Create a search session and return its session_id |
search_papers | Search papers with filters and session deduplication |
search_session_status | Show the progress of one search session |
close_search_session | Close a session and return a short summary |
get_papers | Retrieve papers by OpenReview ID |
list_paper_areas | List available research areas and paper types |
Papers already returned in a session will not appear again. A search without a session_id is not recorded, and sessions do not share state.
2. Add the MCP to an agent
Public MCP endpoint:
https://icml.agent4sci.com:20441/mcpAdd it to Claude Code:
claude mcp add --transport http icml-papers --scope user \
https://icml.agent4sci.com:20441/mcpOther agents with Streamable HTTP MCP support can use:
{
"mcpServers": {
"icml-papers": {
"type": "http",
"url": "https://icml.agent4sci.com:20441/mcp"
}
}
}The public instance is free to try.
3. Source code, documentation, and local deployment
- Source code: github.com/keta1930/icml2026-mcp
- Documentation: English documentation
- Local setup: Local deployment
The project is released under the MIT License. Local deployment requires Python 3.12+, uv, and a DashScope API key with access to qwen3.7-text-embedding. [Note: A private deployment can replace the DashScope embedding service with a local model.]
4. Further development
Further work has two priorities:
- Improve retrieval stability when the same research question is expressed with different wording.
- Add more open scholarly sources after checking their licenses, fields, identifiers, update procedures, and data quality.
Current plans are recorded in the roadmap.
DISCUSSION
Comments