Skip to content

MCP Registry

What it is

The MCP Registry is the central discovery platform and directory for Model Context Protocol (MCP) servers. It provides a standardized way for developers to publish and for users to discover tools that extend the capabilities of AI agents (like Claude Desktop, IDEs, or autonomous scripts).

What problem it solves

Before the registry, MCP implementations were fragmented across GitHub, NPM, and private blogs. The registry addresses this fragmentation by providing a single, authoritative source for discovering publicly available MCP servers. It standardizes server metadata (via server.json), making it easier to find, evaluate, and install tools.

Where it fits in the stack

Automation / Orchestration. It acts as the "app store" or "package manager" equivalent for the AI tool-calling ecosystem. It provides the metadata infrastructure that allows agents to find the right tool for a specific task.

Typical use cases

  • Discovering Integrations: Finding an MCP server that connects Claude to a specific database (e.g., PostgreSQL), service (e.g., Slack), or local tool (e.g., terminal).
  • Evaluating Maturity: Checking the popularity, maintenance status, and exposed toolsets of different MCP implementations before integrating them.
  • Publishing Tools: Providing a standardized way for developers to share their custom MCP servers with the wider community.
  • Protocol Compliance: Verifying that a server follows the official MCP standards for metadata and communication.

Strengths

  • Official Status: Backed by Anthropic and the core MCP development team as the canonical directory.
  • Standardized Metadata: Enforces the use of server.json for consistent tool representation across the ecosystem.
  • Unified Discovery: Creators publish once, and all consumers (IDEs, desktop apps, CLI tools) can reference the same canonical data.
  • Searchable Index: Provides a categorized and searchable interface for finding specific functionalities.

Limitations

  • Discovery Only: Does not host the actual server code or binaries; it provides links to where they are hosted (GitHub, NPM, Docker Hub).
  • Early Stage: The ecosystem is growing rapidly, so quality and documentation of listed servers can vary significantly.
  • Permissions: The registry does not handle authentication or credential management for the servers it lists.

When to use it

  • When you want to see what integrations are available to extend your AI agent's capabilities.
  • When you are developing a new MCP server and want it to be discoverable by other users.
  • When you need to find the correct installation command for an official or community-maintained MCP server.

When not to use it

  • When you are working with private, internal, or proprietary tools that should not be publicly indexed.
  • When using non-MCP-based tool-calling systems (e.g., legacy OpenAI Function Calling).

Getting started

Browsing the Registry

You can browse the registry through the web interface at registry.modelcontextprotocol.io.

Using Registry Metadata in Config

Many MCP clients use the registry metadata to simplify configuration. For example, if a tool is listed on the registry, you can often find the exact claude_desktop_config.json snippet required to run it.

{
  "mcpServers": {
    "everything": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-everything"]
    }
  }
}

Technical examples

Publishing to the Registry

To list a server, developers typically create a mcp-server.json file in their repository root and submit a PR to the official registry repository.

{
  "mcp-server": {
    "name": "my-custom-server",
    "description": "Exposes my custom home automation tools",
    "version": "1.0.0",
    "repository": "https://github.com/user/my-custom-server",
    "license": "MIT"
  }
}

Programmatic Discovery

Advanced clients can fetch the registry index as JSON to provide an "in-app" discovery experience.

# Example of fetching registry metadata (hypothetical endpoint)
curl https://registry.modelcontextprotocol.io/v1/servers/search?q=postgres

Sources / references

Contribution Metadata

  • Last reviewed: 2026-05-14
  • Confidence: high