Smithery occupies a distinct niche: it's not just a directory, it's a hosting platform. You can deploy and run MCP servers directly through Smithery without managing your own infrastructure. For developers who want zero-configuration MCP access, this is a significant differentiator.

The directory is a byproduct of the hosting business: Smithery lists servers that can be deployed on its platform. Coverage (4,000+) is solid for the most popular servers but incomplete relative to the full ecosystem. Ranking is based on Smithery-platform usage signals — servers with more Smithery installs rank higher, regardless of quality or maintenance outside the platform.

• Hosted server deployment — run MCP servers without infrastructure
• One-click install into compatible clients
• API for programmatic access to the registry
• Growing ecosystem of platform-specific tooling

Limitations: Coverage limited to Smithery-compatible deployments. Ranking reflects platform usage, not objective quality signals. Maintenance activity and issue health are not factored into visibility.

Glama applies security scanning and manual review before listing any server. For organizations with strict security requirements — enterprises, regulated industries, teams where installing an unvetted tool is a compliance issue — this is genuinely valuable. It means every listed tool has been looked at by a human and run through automated security checks.

The limitation is inherent to the model: manual review doesn't scale to the speed at which the MCP ecosystem grows. Hundreds of new repositories appear each week; Glama's reviewed list will always represent a small fraction of what's available.

• Security scanning and manual review for every listed tool
• API access to the curated directory
• Strong signal for security-sensitive selection decisions

Limitations: Coverage is limited by manual review capacity. No composite quality score for comparing tools within the vetted set. Ecosystem coverage will always lag broad automated indexes.

MCP.so is a community-run web directory for browsing MCP servers by category. The site aggregates submissions from the community and provides a reasonable browsing interface. There is no composite quality score, no freshness indicator per tool, and no programmatic API. It functions as a search-and-browse tool, not a ranked index.

Coverage at 3,000+ is meaningful but incomplete. MCP.so captures many mainstream servers but misses a large portion of the ecosystem that hasn't been manually submitted. Community-driven means coverage depends on who happens to submit their server.

• Clean browsing interface organized by category
• Community submissions keep popular servers visible
• No login required for browsing

Limitations: No quality scoring. No freshness signals. No API. Coverage depends on manual community submissions rather than automated crawling. Not viable for agentic or programmatic use cases.

mcp-get is a different product category entirely — it's a package manager, not a discovery tool. The command-line interface lets you install MCP servers directly into Claude Desktop and other compatible clients with a single command:

npx @michaellatman/mcp-get@latest install @modelcontextprotocol/server-github

The underlying registry is a curated JSON file (~200 entries) maintained via GitHub PRs. There is no quality scoring, no freshness data, and no discovery UI. mcp-get is most useful as an installation mechanism after you've already decided which server to use.

• One-command installation into Claude Desktop and compatible clients
• Open source package list (JSON on GitHub)
• No manual configuration file editing required

Limitations: Not a discovery tool. 200-entry curated list misses the vast majority of the ecosystem. No scoring, no API, no freshness signals. Intended to complement directories, not replace them.

The official registry at registry.modelcontextprotocol.io is not a developer discovery tool. It's a certification layer — vendors submit MCP servers for official recognition by the MCP team, and the list is curated by Anthropic. At ~87 entries, it covers a tiny fraction of the ecosystem.

What it does represent is a credibility floor. Anything listed here has explicit endorsement from the team that maintains the MCP protocol. For vendors building commercial MCP servers, official listing is a meaningful signal. For developers searching for tools, this list alone will rarely be sufficient.

• Official Anthropic endorsement for listed tools
• High bar for inclusion — meaningful credibility signal
• Best starting point for understanding the reference implementation landscape

Limitations: ~87 entries. No quality scoring within the set. No API for programmatic access. Designed for vendor certification, not developer discovery.