Definition
The Model Context Protocol (MCP) is an open standard, introduced by Anthropic in 2024 and now supported across the industry, that lets AI applications connect to external tools and data through a common interface. An app exposes an MCP server with tools (actions) and resources (data); an AI client such as Claude, ChatGPT, Cursor, or VS Code connects to it and can call those tools on the user's behalf.
Why MCP matters
Before MCP, every AI tool needed a custom integration with every app. MCP is often compared to USB-C: one standard plug, so any compliant AI client works with any compliant server. For a knowledge base, that means you connect your company's knowledge once and every AI tool your team uses can reach it.
How MCP works
- Server: the app with the data, for example Granite, exposes tools like
search_notesorread_note. - Client: the AI application (Claude, ChatGPT, Cursor) discovers those tools and decides when to call them.
- Transport: remote servers use Streamable HTTP over the internet; local servers run on your computer and talk over standard input and output (stdio).
- Authorization: remote servers use OAuth 2.1 so you sign in and approve exactly what the client may do.
Remote vs local MCP servers
| Remote | Local | |
|---|---|---|
| Runs | On the provider's servers | On your computer |
| Best for | Teams, synced data, web AI clients | Private folders, offline work, developers |
| Sign-in | OAuth or a scoped token | None; it's your machine |
| In Granite | https://mcp.granite.md/mcp (Pro, Business) | granite mcp (every plan) |
MCP security in five rules
- Least privilege: grant read before write, and only the vaults or folders needed.
- Per-user identity: connections should act as a person, with that person's permissions.
- Revocability: you should be able to see and revoke every connected client.
- Audit: connections and access should be logged.
- Trusted servers only: install MCP servers from sources you trust; a server can act with the permissions you give it.
Connect your notes in two minutes
- In Claude, go to Settings → Connectors → Add custom connector and enter
https://mcp.granite.md/mcp. In Claude Code:claude mcp add --transport http granite https://mcp.granite.md/mcp. - Sign in to Granite in the window that opens and choose what the tool may do and which vaults it may use.
- Ask: "What did we decide about pricing last month?"
Full instructions for Cursor, VS Code, and local folders are in the MCP setup guide. See everything Granite's MCP server can do.
Frequently asked questions
What does MCP stand for?
MCP stands for Model Context Protocol, an open standard for connecting AI applications to external tools and data sources.
Which AI tools support MCP?
Claude (desktop, web, and Claude Code), ChatGPT through connectors, Cursor, VS Code, Windsurf, Zed, and many agent frameworks support MCP, and the list keeps growing.
Is MCP safe for company data?
MCP is safe when the server enforces per-user permissions, uses OAuth with scoped and revocable access, and logs activity. The protocol carries requests; the server you connect decides what is allowed.
What is the difference between MCP and an API?
An API is built for programmers to call. MCP describes tools in a way AI models can discover and call on their own, with a standard transport and authorization, so one server works with many AI clients without custom code.
