AI4Paper is building an open-source workspace that helps researchers complete the entire paper-writing workflow with AI: discovering and reviewing literature, planning experiments, drafting and revising text, creating figures, managing references, and preparing a paper for submission.
AI4Paper now uses OpenCode as its backend agent runtime, with apaper-plugin providing paper-authoring capabilities and apaper-mcp providing research tools. DSH is no longer the runtime, and development of the former ipaper workspace has stopped.
The repositories form a layered system rather than a collection of separate tools:
flowchart LR
USER((Researcher)) --> OPENCODE
subgraph AGENT [Agent Runtime and Capabilities]
OPENCODE[OpenCode<br/>Backend agent runtime]
PLUGIN[apaper-plugin<br/>Paper-authoring capabilities for OpenCode]
SKILLS[Writing and figure skills]
OPENCODE -->|loads| PLUGIN
PLUGIN --> SKILLS
end
subgraph RESEARCH [Research Services]
MCP[apaper-mcp<br/>Python MCP server]
SOURCES[(Academic paper sources)]
MCP --> SOURCES
end
PLUGIN -->|research tools| MCP
BOOK[book<br/>Paper-writing guide]
BOOK -.->|workflow guidance| USER
BOOK -.->|best practices| AGENT
- OpenCode is the backend agent runtime: it runs the agent that coordinates paper-writing tasks and uses specialized skills and research tools.
apaper-pluginis the agent capability layer: it targets OpenCode and equips the agent with academic writing, figure creation, and paper-research capabilities.apaper-mcpis the research service: a Python MCP server that searches academic sources, retrieves metadata and BibTeX, and downloads papers.bookis the workflow guide: practical guidance and best practices for the complete AI-assisted paper lifecycle.
| Repository | Stack | Role |
|---|---|---|
| apaper-plugin | OpenCode | The academic paper-authoring capability layer for the OpenCode agent runtime. It adds writing and publication-quality figure skills and connects the agent to apaper-mcp research tools. |
| apaper-mcp | Python 3.12+ · uv | A Python MCP server for academic research. It searches arXiv, IACR ePrint, DBLP, Google Scholar, and CNKI; retrieves metadata and BibTeX; and downloads supported PDFs. |
| book | Typst | A practical guide to the full AI/ML paper lifecycle: setup, problem selection, experiments, data and figures, writing, submission, revision, and post-acceptance work. |
| ipaper | Legacy workspace | The former user-facing workspace. Development has stopped; it is not part of the current architecture. |
- A researcher uses OpenCode as the agent runtime for paper-writing tasks.
- OpenCode loads
apaper-pluginto gain specialized writing and figure-generation skills. - The plugin uses
apaper-mcpwhen the agent needs to search the literature, collect citations, or download papers. - The
bookprovides workflow guidance and best practices throughout the process.
For setup and direct use:
- Follow the
apaper-pluginREADME for OpenCode setup instructions. - Run the Python MCP server with
uvx apaper-mcp, or connect any MCP-compatible client to it as a local stdio server.
We welcome contributions across all repositories. Open an issue or pull request in the project most closely related to your change.
See LICENSE for details.