FAQ · numbers updated 2026-10-19
MCP, answered
01MCP basics
What is MCP (Model Context Protocol)?
MCP is an open protocol, introduced by Anthropic in November 2024, that standardizes how AI applications connect to external tools and data. An AI app (the client) connects to MCP servers, and each server exposes tools, resources and prompts the model can use. Claude, ChatGPT, Cursor and VS Code all support it.
What are MCP servers?
An MCP server is a small program that wraps a real system (a database, GitHub, a design tool, a docs site) and exposes it to AI apps over MCP. It either runs locally on your machine over stdio or is hosted at a URL over Streamable HTTP. MCPSkillHub indexes 49,186 MCP servers.
What is an MCP client?
The client is the AI application you talk to, such as Claude Desktop, Claude Code, ChatGPT, Cursor, VS Code or Windsurf. It asks each server what it can do (tools/list), passes those tool definitions to the model, runs the tool calls the model asks for (tools/call), and returns the results to the model.
What are MCP tools, resources and prompts?
Tools are actions the model can call, like creating an issue or querying a database. Resources are data the app can read into context, like a file or a record. Prompts are reusable templates a user can pick. In the MCP specification, tools are model-controlled, resources are application-controlled and prompts are user-controlled.
What is the difference between local and remote MCP servers?
A local MCP server runs as a process on your machine and talks to the client over stdio; it is usually started with npx, uvx or Docker. A remote (hosted) MCP server runs elsewhere and is reached over Streamable HTTP (older servers use SSE), often with OAuth sign-in. 26,290 of the servers we index publish a hosted endpoint.
What is Streamable HTTP in MCP?
Streamable HTTP is the MCP transport for remote servers. It was introduced in the 2025-03-26 revision of the specification and replaced the earlier HTTP+SSE transport. The client sends JSON-RPC messages over HTTP POST, and the server can answer with a single response or stream several messages back.
02Installing and using servers
How do I add an MCP server to Claude Code?
Use the claude mcp add command. For a hosted server: claude mcp add --transport http <name> <url>. For a local one: claude mcp add <name> -- npx -y <package>. Every server page on MCPSkillHub has the exact command to copy, plus configs for Cursor, VS Code, Claude Desktop and Windsurf.
How do I add an MCP server to Cursor or VS Code?
Cursor reads MCP servers from .cursor/mcp.json in your project or ~/.cursor/mcp.json for all projects. VS Code reads them from .vscode/mcp.json in the workspace or from your user settings. Each server page shows the JSON to paste for both.
How do I test or debug an MCP server?
Use the official MCP Inspector (npx @modelcontextprotocol/inspector). It connects to a local or remote server and lets you list and call its tools, resources and prompts by hand, so you can see exactly what a client would see.
What is the official MCP Registry?
The official MCP Registry (registry.modelcontextprotocol.io) is the community registry where maintainers publish server metadata under a verified namespace. MCPSkillHub indexes it (41,266 servers) together with GitHub, npm and PyPI, then adds measurements the registry does not publish.
03Token cost and safety
How many tokens does an MCP server use?
A connected server's tool definitions are sent to the model on every request, unless your client loads tools on demand. Across the 3,254 servers we have measured, the median adds about 1,540 tokens per turn and the 90th percentile adds 23,085. Every server page shows its own count.
How do I reduce MCP token usage?
Connect only the servers a task needs, prefer servers with lean tool lists, turn off tools you don't use if your client allows it, and use a client that loads tool definitions on demand. Prompt caching lowers the price of re-sending the same definitions, but not the context space they take up.
Are MCP servers safe to use?
An MCP server acts with whatever access you give it, so treat it like any third-party code. Use least-privilege, read-only credentials where possible, keep secrets in environment variables rather than committed config, read tool descriptions, prefer maintained servers, run local servers in a sandbox and keep tool approval on. Our security signals list each finding with its evidence; they are signals, not a guarantee.
What do Active, Stale and Abandoned mean on MCPSkillHub?
Freshness comes from a server's latest activity across repository commits, releases, package publishes and registry updates. Active means activity within 30 days, Maintained within 90, Slowing within 180, Stale within a year, and Abandoned longer than that. Deprecated means the maintainer marked it so.
04About MCPSkillHub
Is MCPSkillHub free, and is it affiliated with Anthropic?
MCPSkillHub is free and independent. It is not affiliated with Anthropic, the Model Context Protocol project or the official MCP Registry. We index public sources, link back to them and publish our own measurements.
How do I submit or claim my MCP server?
Submit any public server at mcpskillhub.com/submit. To claim a listing and get the verified badge, sign in and verify either push access to its GitHub repository or a DNS TXT record on its domain. Claimed listings can request re-analysis.
Can I use MCPSkillHub's data?
Yes. The context-tax dataset is free under CC BY 4.0 as a CSV, each server has JSON at https://mcpskillhub.com/api/v1/servers/{slug}, and our own remote MCP server at https://mcpskillhub.com/mcp offers search_servers, get_server and compare_servers to any MCP client.
Something missing or wrong? Email admin@mcpskillhub.com.