How We Built Our MCP Server Registry

The MCP Ecosystem Problem

The Model Context Protocol ecosystem has exploded in 2026. There are now hundreds of MCP servers available, but discovering them is a challenge. They are scattered across GitHub repositories, npm packages, and individual blogs with no centralized directory.

We set out to build the most comprehensive and reliable MCP server registry on the web. Here is how we did it.

Data Sourcing Pipeline

Our registry pulls data from multiple sources:

GitHub API — We scan for repositories with MCP-related topics and keywords, extracting metadata like star counts, last update times, and contributor information

npm Registry — We monitor npm for packages matching MCP server patterns, tracking weekly downloads and version history

Manual Curation — Our editorial team reviews submissions and verifies claims about compatibility and features

The data pipeline runs on a scheduled basis, with GitHub data refreshed daily and npm data hourly. We use a simple queue-based system built on top of our existing infrastructure.

Verification System

Not all MCP servers are created equal. We built a multi-tier verification system:

Automated checks — We verify that the GitHub repository exists, the npm package installs without errors, and basic configuration files are present

Compatibility testing — We test each server against Claude Desktop, VS Code, and Cursor to verify claimed compatibility

Security review — Our team reviews the code for obvious security concerns and documents any required permissions

Verified servers get a badge in the registry, giving developers confidence in what they are installing.

The Schema Design

We store MCP server data in our SQLite database using Prisma. The schema includes fields for transport type (stdio vs SSE), compatible clients, feature lists, and security notes. We store array data as JSON strings to keep the SQLite schema simple while maintaining query flexibility.

Search and Discovery

Finding the right MCP server means filtering by multiple dimensions simultaneously. We built a faceted search that lets developers filter by category, transport type, compatible client, and verification status. The search uses database indexes for fast filtering and returns results ordered by a composite relevance score that factors in GitHub stars, download counts, and verification status.

Lessons Learned

Building the registry taught us several lessons. First, data freshness matters enormously — developers want to know if a project is actively maintained. Second, compatibility claims need verification because many servers only work with specific clients. Third, having an editorial layer on top of automated data collection adds significant trust.

The MCP registry is one of our most visited sections, and we are continuously improving it based on community feedback.