Motivation
The alphaXiv MCP already provides a strong set of primitives for agentic research workflows: discovering papers, reading papers deeply, inspecting the user's library, and organizing papers into folders.
One important part of the reading workflow is currently inaccessible through MCP: highlights and annotations created while reading papers in alphaXiv.
Exposing these would allow an agent to work with the user's accumulated understanding of a paper, rather than treating every paper as if it were being encountered for the first time.
Use case
I use alphaXiv as part of a research workflow running through ChatGPT's hosted environment.
When I read a paper myself, the passages I highlight and the annotations I make are often more valuable to a downstream agent than the full paper alone. They encode what caught my attention, what I considered important, and sometimes my own interpretation or questions.
If these were available through MCP, an agent could:
- retrieve my highlights and annotations for a paper
- distinguish my reading state from the underlying paper content
- use my previous observations when revisiting a paper
- synthesize ideas across papers based partly on what I found important
- preserve continuity between human reading and agent-assisted research
- incorporate that accumulated context into recurring research and knowledge workflows
Without this surface, the paper itself is accessible to the agent, but an important layer of the user's interaction with that paper is lost.
Proposed capability
Expose MCP tools for retrieving a user's highlights and annotations for a paper.
Useful fields could include:
- paper ID
- highlight ID
- highlighted text
- location / page where available
- annotation or note
- creation and modification timestamps
Ideally, the MCP would eventually support read and write access, including:
- creating a highlight
- attaching an annotation or note
- updating an annotation
- deleting a highlight or annotation
Read access alone would already unlock substantial value, so write support does not necessarily need to block an initial implementation.
Why this matters
Highlights and annotations form a useful bridge between paper content and personal research context.
For an agentic research workflow, this enables a useful loop:
discover → read → capture → organize
The alphaXiv MCP already covers much of the reading and organization side. Exposing reading-state data would allow agents to participate in the same evolving research context as the human reader rather than operating on the paper corpus independently.
This becomes especially useful for recurring workflows where papers are revisited over time.
Related MCP capability
This complements #377, which proposes exposing the alphaXiv For You and Hot feeds through MCP.
The feed capability provides the discovery side of the workflow, while highlights and annotations preserve the user's reading and interpretation state.
Together, the two capabilities would make it possible for hosted agents to work across much more of the alphaXiv research lifecycle:
discover → read → capture → organize
Motivation
The alphaXiv MCP already provides a strong set of primitives for agentic research workflows: discovering papers, reading papers deeply, inspecting the user's library, and organizing papers into folders.
One important part of the reading workflow is currently inaccessible through MCP: highlights and annotations created while reading papers in alphaXiv.
Exposing these would allow an agent to work with the user's accumulated understanding of a paper, rather than treating every paper as if it were being encountered for the first time.
Use case
I use alphaXiv as part of a research workflow running through ChatGPT's hosted environment.
When I read a paper myself, the passages I highlight and the annotations I make are often more valuable to a downstream agent than the full paper alone. They encode what caught my attention, what I considered important, and sometimes my own interpretation or questions.
If these were available through MCP, an agent could:
Without this surface, the paper itself is accessible to the agent, but an important layer of the user's interaction with that paper is lost.
Proposed capability
Expose MCP tools for retrieving a user's highlights and annotations for a paper.
Useful fields could include:
Ideally, the MCP would eventually support read and write access, including:
Read access alone would already unlock substantial value, so write support does not necessarily need to block an initial implementation.
Why this matters
Highlights and annotations form a useful bridge between paper content and personal research context.
For an agentic research workflow, this enables a useful loop:
discover → read → capture → organize
The alphaXiv MCP already covers much of the reading and organization side. Exposing reading-state data would allow agents to participate in the same evolving research context as the human reader rather than operating on the paper corpus independently.
This becomes especially useful for recurring workflows where papers are revisited over time.
Related MCP capability
This complements #377, which proposes exposing the alphaXiv For You and Hot feeds through MCP.
The feed capability provides the discovery side of the workflow, while highlights and annotations preserve the user's reading and interpretation state.
Together, the two capabilities would make it possible for hosted agents to work across much more of the alphaXiv research lifecycle:
discover → read → capture → organize