MCP.so Server Directory Scraper

Scrape the complete MCP.so server directory — the 3rd canonical MCP registry alongside Smithery and Glama. Extracts name, author, category, tags, capabilities, tool count, install command, GitHub repo URL, README markdown, and more for every listed MCP server.

What does it scrape?

MCP.so is one of the top-3 community indexes for Model Context Protocol (MCP) servers — the protocol used by Claude Desktop, Cursor, Windsurf, and Codex to connect AI models to external tools. This scraper enumerates every server in the directory and returns a structured record per entry.

Data fields returned

Field	Description
server_id	Unique identifier in <slug>/<author> format
server_name	Display name of the MCP server
server_slug	URL slug
author_username	GitHub username of the author
author_avatar_url	Author avatar image URL
tagline	Short description
description	Full description
category	Category (databases, developer-tools, etc.)
tags	Comma-separated tags
capabilities	MCP capability types present (tools, resources, prompts)
tool_count	Number of MCP tools exposed
install_command	npx / uvx / docker installation command
npm_package	NPM package name (if applicable)
pypi_package	PyPI package name (if applicable)
docker_image	Docker image name (if applicable)
github_repo	GitHub repository URL
is_official	Whether this is an official MCP team server
is_verified	Whether marked as featured/verified
listed_at	Date first listed on mcp.so
updated_at	Last update timestamp
readme_markdown	Full README content
profile_url	URL of the mcp.so server page
scraped_at	Scrape timestamp

Usage

Set maxItems to limit the number of servers returned. The directory has ~2,000 servers across two sitemaps.

Input

{
  "maxItems": 100
}

Output sample

{
  "server_id": "redis/modelcontextprotocol",
  "server_name": "Redis",
  "server_slug": "redis",
  "author_username": "modelcontextprotocol",
  "category": "databases",
  "tags": null,
  "capabilities": "tools",
  "tool_count": 13,
  "install_command": "docker run -i --rm mcp/redis {redis_url}",
  "docker_image": "mcp/redis",
  "github_repo": "https://github.com/modelcontextprotocol/servers/tree/main/src/redis",
  "is_official": false,
  "is_verified": false,
  "listed_at": null,
  "profile_url": "https://mcp.so/servers/redis"
}

Triple MCP Coverage

This scraper completes our 3-index MCP directory coverage:

Registry	Actor
Smithery	jungle_synthesizer/smithery-mcp-registry-scraper
Glama	jungle_synthesizer/glama-mcp-registry-scraper
MCP.so	This actor

Use all three to build an exhaustive MCP server catalog for agent platform development, ecosystem analysis, or security research.

Notes on data availability

mcp.so migrated its front-end to TanStack Start (2026-07, FR 480), which changed both the sitemap shape (/sitemap.xml index → ?section=servers&page=N children) and the detail-page URL shape (/server/<slug>/<author> → /servers/<slug>). Structured data is now sourced from the page's JSON-LD SoftwareApplication block plus the rendered install-config code block and tools accordion, rather than the old embedded RSC payload.

As a result the following fields are no longer exposed by mcp.so and will always be null/false in every record: tags, resource_count, prompt_count, github_stars, github_forks, language, license, view_count, install_count, listed_at, updated_at, documentation_url, is_official, is_verified (is_official/is_verified are always false — there is no longer a reliable official/verified signal on the detail page). github_stars, github_forks, language, and license would require GitHub API enrichment as a separate step.