Hugging Face's logo

Datasets:
CyberMax-tools
/
toolroll-mcp-registry-snapshot

Tasks:
Text Classification
Text Retrieval
Modalities:
Text
Tabular
Formats:
parquet
Languages:
English
Size:
10K - 100K
Tags:
mcp
model-context-protocol
mcp-servers
ai-agents
agent-tools
tool-use
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
toolroll-mcp-registry-snapshot / README.md
CyberMax-tools's picture
CyberMax-tools
Toolroll brand kit: banner + gallery
1f2aeaf verified
preview code
|
Raw History Blame
6.07 kB
metadata 
license: other
license_name: mcp-registry-public-metadata
license_link: >-
  https://huggingface.co/datasets/CyberMax-tools/toolroll-mcp-registry-snapshot#source-and-licence
pretty_name: 'Toolroll: Official MCP Registry Snapshot (every MCP server, weekly)'
language:
  - en
tags:
  - mcp
  - model-context-protocol
  - mcp-servers
  - ai-agents
  - agent-tools
  - tool-use
  - llm-tools
  - registry
  - tabular
  - cybermax
size_categories:
  - 10K<n<100K
configs:
  - config_name: servers
    default: true
    data_files:
      - split: train
        path: data/servers.parquet
task_categories:
  - text-classification
  - text-retrieval

Toolroll: every server in the official MCP Registry, as one clean table

Toolroll: Every MCP server in one table

Try the paid tool: Capturewren on Apify: give your agent eyes: screenshot + page text as an MCP tool. First try costs cents; pay only for results.

A snapshot of the official Model Context Protocol (MCP) Registry (registry.modelcontextprotocol.io) as a single Parquet table: 35,403 MCP servers (35,028 active, 375 deprecated) from 20,000+ publishers, snapshot 2026-09-24 07:03 UTC. One row per server (its latest version) with its description, how you connect to it (remote URL and transport, or npm / PyPI / Docker (OCI) / MCPB / NuGet / Cargo package), whether it needs a secret (API key header or secret env var), repository, website and publish dates.

Use it to find MCP tools for an agent, study what gets built, track the ecosystem week by week, or seed a tool-search index. The registry API is paginated and version-heavy (many servers re-publish hundreds of versions); this gives you the current state in one 4 MB file.

Toolroll: product image 1 Toolroll: product image 2 Toolroll: product image 3

What's in this snapshot (active servers)

Servers
Active servers 35,028
Reachable as a hosted remote (no install) 21,428 (streamable-http 20,693, SSE 1,066)
Installable package 14,903 (npm 9,624, PyPI 3,870, OCI 954, MCPB 879, NuGet 128, Cargo 57)
Need a secret (API key header or env var) 8,436
Latest version published in the last 30 days 13,734

A handful of namespaces publish thousands of generated servers each, so filter by namespace if you want hand-built tools only.

Load it 

from datasets import load_dataset
ds = load_dataset("CyberMax-tools/toolroll-mcp-registry-snapshot", split="train")
remote_no_key = ds.filter(lambda r: r["status"] == "active" and r["has_remote"] and not r["needs_secret"])

import pandas as pd
df = pd.read_parquet("https://huggingface.co/datasets/CyberMax-tools/toolroll-mcp-registry-snapshot/resolve/main/data/servers.parquet")
df[df.description.str.contains("geocod", case=False)][["name", "description", "remote_urls"]]

Columns

name (registry id, reverse-DNS namespace/name), namespace, title, description, version, status (active/deprecated), is_latest, published_at, updated_at (of this version), website_url, repository_url, repository_source, package_registries, package_identifiers, package_transports, remote_types, remote_urls, has_remote, has_package, env_var_count, needs_secret.

MCP tools from CyberMax

Pay-per-call data tools any MCP client can call through Apify's hosted MCP server (no install; bring an Apify token):

  • Insidewell: SEC Form 4 insider trades with cluster-buy flags
  • DomainDNA: tech stack, email security (SPF/DMARC) and domain age for any domain
  • HireHeat: which companies are hiring, for what, by domain
  • Capturewren: website screenshots plus page text for AI agents
  • Swellmeter: Google Trends for AI agents

Connect: https://mcp.apify.com/?tools=cybermax/<tool> (e.g. cybermax/domain-intel).

When NOT to use this

  • You need the tool-level schemas (each server's tool names and input schemas): the registry stores server metadata only; tools are listed by the server itself at runtime.
  • You need quality or uptime signals: registry presence says nothing about whether a server works.

Source and licence

Collected from the public official MCP Registry API (GET /v0/servers?version=latest). Server names, descriptions and links are metadata published by each server's author to a public registry; the compilation is shared here for discovery and research. Each server's own code and content keep their own licences. Entries from the publisher's own namespace are omitted. Not affiliated with or endorsed by the MCP project. Refresh: weekly.

Published by CyberMax.

More free from CyberMax

Quickstart notebook

Load this dataset with pandas and see real outputs in the quickstart notebook.