Annota is an AI-driven paper annotation MCP server, aimed at researchers who use Zotero to manage literature. It lets AI read PDF papers, understand the content, generate precise annotations, and save the results back to the reference manager.
Main capabilities: AI reads papers and creates annotations; generates structured reading summaries containing key findings, methods, and conclusions; provides 3 Claude Code skills (slash commands); uses a two-stage workflow, first reading the full text (low cost), then fetching only the coordinates of the target sentence (precise), which can reduce context usage by 63–80%; automatically detects and skips the "References" section, for example extracting only 13 pages from a 21-page paper; batch annotation, creating 10 highlights in a single API call; returns friendly errors instead of crashing when writes fail.
How to use: first clone and install, then configure it in ~/.claude.json or Claude Code's MCP Servers settings. Zotero data directory locations: on Windows, Zotero → Edit → Settings → Advanced → Data Directory Location (default C:\Users\YourName\Zotero); on macOS, Zotero → Settings → Advanced → Data Directory Location (default ~/Zotero); on Linux, the default is ~/Zotero. After that, you can directly converse with Claude in natural language, or use slash commands.
Applicable scenarios: highlighting key findings, annotating assumptions and theories, explaining formulas, recording policy implications and conclusions, generating reading notes for an entire paper, detailed notes paragraph by paragraph, etc.
Limitations and disclaimer: Annota writes directly to Zotero's SQLite database, bypassing Zotero's internal consistency mechanisms; this is a design choice for achieving a fully offline, local-first annotation workflow. Users are responsible for the database themselves; please back up zotero.sqlite before use. Future versions plan to migrate to the official Zotero Web API / Local API. The project uses the MIT license and can be used for research and commercial projects.
