Frequently asked questions
Short, direct answers — for more detail on how MCPstack actually works, see About.
What is MCP (Model Context Protocol)?
MCP is an open protocol that lets an AI agent connect to external tools and data — files, APIs, databases — in a standard way, instead of every integration being custom-built.
What is an MCP server?
An MCP server is a program that exposes a specific set of tools or data over MCP — a GitHub server exposes repos and issues, a database server exposes queries, and so on. An agent connects to one to gain that capability.
What's the difference between servers, clients, CLI tools, skills, and agents on MCPstack?
Servers expose capabilities to an agent. Clients are the applications that host an agent and connect to servers (Claude Desktop, Claude Code). CLI tools are for building or operating MCP servers directly. Skills are packaged instructions for one task, not a full server. Agents are pre-built configurations for a specific job.
Is it safe to install an MCP server?
Treat an MCP server like any other third-party code you'd run — check what it actually does before granting it access, especially anything with filesystem, network, or credential access. MCPstack's review confirms a listing is real and accurately described; it isn't yet a security audit of every listing (see the About page for what review does and doesn't cover today).
What does the "Official" badge mean?
It means the listing is published directly by the vendor's own organization — Microsoft's own repo, not a third-party reimplementation. It's a provenance signal, separate from hands-on verification. More detail on the About page.
How do I submit a listing?
Use the submission form. It goes into a moderation queue and gets reviewed before it's published — nothing goes live automatically.
How often is the catalog updated?
New listings are discovered automatically on a schedule via GitHub and npm, and every one sits in the moderation queue until reviewed and approved.