For agents

Query all of this from your agent

Everything on this site is exposed over the Model Context Protocol at a single HTTP endpoint. No key, no signup, no rate limit worth mentioning. The underlying data is public JSON anyway, so a token would add friction without adding protection.

Connect

Claude Code:

claude mcp add --transport http robotics-papers https://robotics-papers.gtmascode.dev/api/mcp/

Anything that reads a standard MCP config:

{
  "mcpServers": {
    "robotics-papers": {
      "type": "http",
      "url": "https://robotics-papers.gtmascode.dev/api/mcp/"
    }
  }
}

Note the trailing slash: it is the canonical path, and without it you get a redirect your client has to follow. A plain GET there returns the tool list, which is a quick way to check it is reachable before wiring anything up.

Tools

Tool	What it returns
`search_papers`	Search 1,000+ arXiv cs.RO papers by text, topic, date, citations, or whether a club has presented them.
`get_paper`	One paper in full, with its six-slide presenter kit and the sessions that presented it.
`search_repos`	Robotics repositories with their creator, filterable by owner, organisation, or whether a lab or one person is behind it.
`list_chapters`	Reading clubs that meet in public, plus the cities that have labs and no club.
`list_events`	Sessions and their pre-reading lists, each resolved to its arXiv record.
`search_researchers`	People who have presented, their stated role, and their public links.
`recommend_papers`	Similar papers for a given id, or the ranked list of papers nobody has presented yet.
`get_trends`	Each area's share of monthly cs.RO over 24 months, with the arXiv query behind every number.
`list_topics`	The taxonomy: slug, label, description, current share.

What is behind it

2,361 papers, 262 repositories, 13 reading clubs, 38 researchers, and 24 months of topic trend data. Refreshed twice a week, and the accuracy of each claim class is re-checked against live sources on every refresh.

Things worth asking it

"Find me a robot-learning paper from the last two months that a club has already presented, on tactile sensing, and give me the slide kit." : search_papers then get_paper.
"Which robotics areas are gaining share on arXiv, and what is the query behind that number?" : get_trends returns the arXiv query string with every series, so the agent can verify rather than trust.
"Is there a reading club in Zurich, and when does it meet?" : list_chapters.
"Show me repos an author actually linked from their own paper, not ones merely cited." : search_repos with linked_to_paper; the link_strength field distinguishes an endorsement from a passing mention.

One caveat worth passing to your agent

Reading-list entries and arXiv titles frequently disagree, because clubs use nicknames and descriptions rather than the published title. list_events returns both listed_as and arxiv_title for exactly this reason. If your agent is matching on titles, match on the arXiv one.

Prefer raw files? Every dataset is at /data as plain JSON, CORS open.