Model Context Protocol

ClusterFunk inside Claude

Connect the ClusterFunk MCP server and Claude can read this month's research trajectories directly — 36 tracked topics across arXiv and bioRxiv, their velocity, maturity and attention gaps, and what is converging with what. It is free, needs no API key, and is read-only.

Endpoint

https://clusterfunk.co/api/mcp

Streamable HTTP, JSON-RPC 2.0. Protocol version 2025-06-18. No authentication.

Installing it

Claude Code

One command. Run it in any project, or add --scope user to make it available everywhere.

claude mcp add --transport http clusterfunk https://clusterfunk.co/api/mcp

Claude desktop and web

Open Settings → Connectors → Add custom connector, then paste the endpoint above. Claude will list the four tools once it connects.

Any other MCP client

Standard HTTP server entry — no wrapper or proxy process needed.

{
  "mcpServers": {
    "clusterfunk": {
      "type": "http",
      "url": "https://clusterfunk.co/api/mcp"
    }
  }
}

What it gives Claude

list_tracked_topics

Every topic ClusterFunk tracks, grouped by category, with slugs. Useful first call when you name a field loosely and Claude needs to know whether it is covered.

Ask Claude: “What research areas does ClusterFunk track?”

get_topic_trajectory

Paper volume, research maturity, velocity rank against every other tracked topic, 3-month direction, and the gap between research activity and public search interest — with explicit caveats wherever the model cannot support a firm number.

Ask Claude: “What's the trajectory on solid-state batteries?”

find_convergences

What is converging with a given topic, and how strong the evidence is: detected (a measured citation link above threshold), inferred (real similarity below it), or emerging (below threshold but rising on a statistically significant trend).

Ask Claude: “What's converging with neuromorphic computing?”

list_emerging_convergences

This month's emerging pairs — overlap still below the documented-link threshold, but climbing. Some months this is empty, which is a real result rather than a gap.

Ask Claude: “What's quietly starting to converge this month?”

What it does and doesn't include

The MCP server serves the free tier: topic-level figures for the current issue (October 2026). That is the same data an anonymous visitor sees on the public pages — nothing behind the paywall is exposed through it.

Included, free

Topic-level maturity, velocity and direction
Attention gap against public search interest
Convergence evidence tiers per topic
This month's emerging convergences

On the site, for subscribers

Sub-cluster breakdowns and adoption curves
The month-by-month archive
Full convergence roadmap with horizons
Watchlists and due-diligence reports

Limits worth knowing

Current month only. The tools always read the latest issue. There is no way to query a past month over MCP; the archive lives on the site.
30 requests a minute. Comfortable for a conversation, deliberately not comfortable for a scraper. Over the limit you get a JSON-RPC error and a Retry-After.
Read-only, and it stays that way. No tool writes anything. The server holds no session state and sets no cookies.
Percentages are floors. An adoption figure is only identifiable once a field's publication rate has peaked, and most tracked clusters have not. The tools say so in-line rather than handing Claude a confident number to repeat. How the method works.

Prefer to browse it yourself?

Open the radar Convergence roadmap