Ask CGIAR Research
A public tool that lets AI assistants search openly licensed CGIAR research publications and answer agrifood questions with citations to the source documents.
It is a Model Context Protocol (MCP) server, so it works
with Claude and any other assistant or agent framework that supports remote MCP servers. Assistants
get two tools: search_cgiar_publications (hybrid keyword + semantic search with country,
region, year and language filters) and get_cgiar_document (read the passages around a hit).
Connect
Server URL: https://cgiar-ask-cgiar-research.hf.space/mcp (no sign-in).
Claude Code: plugin with search guidance (recommended)
claude plugin marketplace add cgiar/ask-cgiar-research
claude plugin install ask-cgiar-research@cgiar
The plugin adds the server and a skill that teaches the assistant how to search, read and cite the
evidence. Then ask, for example: /ask-cgiar-research:ask How do smallholders cope with drought in the Sahel?
Claude Code: server only
claude mcp add --transport http cgiar-research https://cgiar-ask-cgiar-research.hf.space/mcp
Claude.ai and Claude Desktop
Customize → Connectors → Add custom connector, paste the server URL, and choose no sign-in. For better answers, also upload the skill (ask-cgiar-research-skill.zip) under Customize → Skills.
Other assistants and agent frameworks
Add the server URL as a remote MCP server using streamable HTTP transport. The server sends usage guidance to clients that support MCP server instructions.
What is included
Publications from the GARDIAN-indexed CGIAR research outputs that carry an open license allowing redistribution: CC-BY (27,771), CC-BY-SA (1,630) and CC0 (155). About 84% are full text; the rest are abstracts. Tables and figure text were removed during processing, so check the source document for exact figures. Coverage is uneven: if nothing is found, that is not evidence that no CGIAR research exists on the topic.
Privacy and usage data
To report on use, CGIAR records for each tool call: the time, the tool, the search query after automated removal of personal information (names, emails, phone numbers, URLs, ID numbers and precise coordinates), filters, the documents returned, response time, the name of the client application, and a session identifier that changes daily. IP addresses and raw queries are not stored. Records are kept in a private CGIAR repository for 24 months and then deleted. Automated removal is not perfect, so please do not include personal details in questions.
Licensing
Each result shows its document's license and source URL. Reuse of passages must follow that license (attribution for CC-BY; attribution and share-alike for CC-BY-SA). The server code is released under the Apache License 2.0.
Contact
Questions and feedback: Jawoo Koo, via the IFPRI contact form.