BIFF-AI/mcptoplist

MCP Ecosystem Dataset

A daily-refreshed, research-friendly snapshot of the Model Context Protocol (MCP) server ecosystem, published by mcptoplist.com. It covers every server tracked across the five major MCP registries — the Official MCP Registry, Glama, Smithery, mcp.so, and PulseMCP — deduplicated to canonical entities, with per-registry membership and daily registry growth series.

• Current version: v2026-11-09 (generated 2026-11-09)
• License: CC BY 4.0 — free to use with attribution to mcptoplist.com
• Refresh cadence: daily (~04:30 UTC, after the overnight registry syncs settle); each day is one commit, so any past day is retrievable from the repo history
• Collection methodology: https://mcptoplist.com/methodology and https://mcptoplist.com/sources

What is MCP?

The Model Context Protocol is an open standard that lets AI assistants connect to external tools and data sources. Since its launch in November 2024 an ecosystem of tens of thousands of MCP servers has grown across several competing registries. This dataset measures that ecosystem: which servers exist, where they are listed, and how each registry has grown over time.

Files

servers.parquet — 176,933 rows, one per canonical server

Registries list the same server under different names and keys; rows here are canonical entities after cross-registry deduplication (repository-URL and package-identifier resolution, plus rename tracking).

Bulk-listing floods are left out, here and in registry_daily.parquet: an official-registry namespace that lists 200+ servers, averaging 50+ per repository, nearly all remote and all served from at most two hosts (one deployment published under thousands of names). So are repository clone farms: bursts of zero-star, zero-fork copies of an existing repository name on a crawler registry (Glama), listed by no other registry and backed by no package, where the copies outnumber the name's established owners (2026-10: throwaway GitHub accounts re-uploading existing MCP servers, ~5.5k listings). Both rules are mechanical and are the same ones the site applies, so dataset and leaderboard agree on which servers exist; a listing returns as soon as it gains a star, a fork, a package or a second registry.

Column	Type	Description
canonical_server_key	string	Stable canonical key (srv_ + hash). Joins across dataset versions.
name	string	Server name from the canonical listing.
repo_url	string	Canonical source repository URL, when known.
on_official_mcp_registry	bool	Ever listed on the Official MCP Registry.
on_glama	bool	Ever listed on Glama.
on_smithery	bool	Ever listed on Smithery.
on_mcp_so	bool	Ever listed on mcp.so.
on_pulsemcp	bool	Ever listed on PulseMCP.
first_seen	date	Earliest date we observed the server on any registry.
is_active	bool	Listed by at least one registry at generation time under the site's liveness rule (delta-synced registries: not reported deleted; crawled registries: seen within 30 days of that registry's newest crawl). Delisted servers are kept with is_active = false and no rank; mcptoplist.com counts active servers only.
transport_classification	string	stdio / remote transport class, when known.
setup_classification	string	runs_out_of_the_box, needs_credentials, or hosted_remote_auth; null when unknown.
repo_stars_current	int	GitHub stars, snapshot at generation time.
mcptoplist_rank	int	The server's position in the mcptoplist.com ranking (1 = top), exactly as published on the site at generation time; null for servers the site does not rank. The underlying score, its components and weights are not part of the dataset — see https://mcptoplist.com/methodology.

registry_daily.parquet — 990 rows, one per source × day

Column	Type	Description
source	string	One of official_mcp_registry, glama_ai, smithery, mcp_so, pulsemcp.
date	date	Calendar day (UTC); the series ends at the last complete day before generation.
listed_servers	int	Cumulative count of distinct servers whose listing on this source began on or before this day.

manifest.json

Version label, generation timestamp, row counts, and column lists for the release.

Citation

If you use this dataset, please cite mcptoplist.com:

@misc{mcptoplist_dataset,
  title        = {MCP Toplist: MCP Ecosystem Dataset},
  author       = {{MCP Toplist}},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/BIFF-AI/mcptoplist}},
  note         = {Data from \url{https://mcptoplist.com}, CC BY 4.0}
}

Or in prose: “MCP Ecosystem Dataset, mcptoplist.com (CC BY 4.0), https://mcptoplist.com”.

Maintainer

Published by MCP Toplist (BIFF.ai). Questions and issues: hello@biff.ai, or the dataset's community tab.