Motivation
The alphaXiv MCP already provides a strong research surface for agents: paper discovery, deep paper reading, library access, and folder organization.
One important piece is currently missing: access to the same For You and Hot feeds available in the alphaXiv app.
For agentic research workflows, these feeds are significantly more valuable than simply running another recency-based paper search. They contain signals alphaXiv has already computed from personalization, community activity, and relevance.
Use case
I use alphaXiv as part of a scheduled research workflow running inside ChatGPT's hosted environment.
In a local coding harness, I could call alphaXiv APIs or third-party clients directly. In ChatGPT Tasks / hosted workflows, however, the agent is limited to authenticated integrations exposed through MCP.
If the MCP exposed the For You and Hot feeds, an agent could:
- inspect the same daily research stream I see in alphaXiv
- evaluate papers against my current research interests
- deeply read only the papers that clear a relevance threshold
- compare them against papers already present in my library
- organize useful papers into alphaXiv folders
- produce a compact research digest without recreating alphaXiv's recommendation layer independently
This would make the MCP much more useful as a persistent research-discovery substrate rather than only a search interface.
Proposed capability
Expose MCP tools for retrieving:
Useful optional fields/controls would include:
- pagination / cursor
- paper ID
- title and authors
- recommendation/feed rank
- publication date
- alphaXiv URL
- engagement or popularity signals where available
- whether the paper is already in the user's library
- any recommendation reason or personalization metadata that can safely be exposed
Even a minimal endpoint returning the same ordered paper IDs shown in the app would already be extremely useful.
Why this matters
A generic discover recent papers about X query can approximate topical discovery, but it cannot reproduce the personalized and community-derived signal already present in alphaXiv's feed.
Making that signal available through MCP would let research agents build on alphaXiv's recommendation system instead of redundantly rebuilding it.
For users running recurring research workflows through hosted agent environments, this is a particularly strong multiplier.
Related MCP capability
This complements #378, which proposes exposing highlights and annotations through the alphaXiv 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 support a broader agentic research loop:
discover → read → capture → organize
Motivation
The alphaXiv MCP already provides a strong research surface for agents: paper discovery, deep paper reading, library access, and folder organization.
One important piece is currently missing: access to the same For You and Hot feeds available in the alphaXiv app.
For agentic research workflows, these feeds are significantly more valuable than simply running another recency-based paper search. They contain signals alphaXiv has already computed from personalization, community activity, and relevance.
Use case
I use alphaXiv as part of a scheduled research workflow running inside ChatGPT's hosted environment.
In a local coding harness, I could call alphaXiv APIs or third-party clients directly. In ChatGPT Tasks / hosted workflows, however, the agent is limited to authenticated integrations exposed through MCP.
If the MCP exposed the For You and Hot feeds, an agent could:
This would make the MCP much more useful as a persistent research-discovery substrate rather than only a search interface.
Proposed capability
Expose MCP tools for retrieving:
Useful optional fields/controls would include:
Even a minimal endpoint returning the same ordered paper IDs shown in the app would already be extremely useful.
Why this matters
A generic
discover recent papers about Xquery can approximate topical discovery, but it cannot reproduce the personalized and community-derived signal already present in alphaXiv's feed.Making that signal available through MCP would let research agents build on alphaXiv's recommendation system instead of redundantly rebuilding it.
For users running recurring research workflows through hosted agent environments, this is a particularly strong multiplier.
Related MCP capability
This complements #378, which proposes exposing highlights and annotations through the alphaXiv 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 support a broader agentic research loop:
discover → read → capture → organize