Guide

What is MCP (Model Context Protocol)?

MCP — the Model Context Protocol — is an open standard which lets an AI assistant connect to real apps and take actions in them, rather than only chatting. Anthropic introduced and open-sourced MCP in late 2024 and published the specification, so any AI client or application can implement the protocol. MCP defines a common way for AI clients — Claude, Claude Desktop, ChatGPT in developer mode, Cursor, Codex CLI — to talk to servers, which are the apps themselves. Because the protocol is shared rather than vendor-specific, one server works across every compatible client. Additionally, a client adding MCP support immediately reaches every published server. Before MCP, each AI-to-app integration was written from scratch and worked in exactly one place. MCP server meaning, in plain English: an app which exposes data and actions to an AI assistant. In practice, the sections below show how the pieces fit, plus a real server you can connect with Claude in about a minute.

What does "MCP server" mean?

An MCP server is an application which exposes data and actions to AI assistants through the Model Context Protocol. The assistant connects to the server once, then reads information and performs real operations — creating, updating and deleting records in a live account, rather than describing an intention. For example, a file server lets an assistant read documents, a calendar server lets an assistant create events, and a planner server lets an assistant plan. Salta's MCP server, at https://mcp.getsalta.app, publishes 18 tools covering pure CRUD over tasks and goals. Each tool is scoped to one account and carries a risk hint, so the client knows before calling whether an action reads or writes. Additionally, no intelligence lives in the server: the assistant does the reasoning, calls the tools, and the user approves whatever lands. In practice, an MCP server is the door an app opens for Claude, Cursor and every other client speaking the protocol.

How do MCP clients and servers work together?

MCP has exactly two sides: a client, which is the AI app you talk to, and a server, which is the app exposing data and actions over the protocol. First, you connect a client to a server once, by pasting the server's URL into the client's connector settings. Second, you approve the connection. From then on the assistant uses that server's capabilities inside an ordinary conversation — no plugin store, no bespoke integration, no separate window. The Model Context Protocol matters here precisely because the standard is shared: build one server and every compatible client reaches the server; add MCP support to one client and the client reaches every published server. For example, Salta's server authenticates with OAuth 2.1, so you sign in to a real account instead of pasting an API key into a config file. In practice, the client asks, the server answers, and you stay in control of whatever actually saves.

Claude MCP servers: what can you connect?

Claude MCP servers are simply MCP servers connected to Claude, Claude Desktop or Claude Code — and because the protocol is not vendor-specific, the same servers also work with Cursor, Codex CLI and ChatGPT's developer mode. That portability is the point of a standard: nobody buys into one assistant's plugin ecosystem. Additionally, public directories index what is available — the Official MCP Registry at registry.modelcontextprotocol.io, plus Smithery, Glama, PulseMCP and mcp.so. For example, Salta is published on the Official Registry as app.getsalta/salta and scores 98/100 on Smithery's quality scan. Connecting takes two steps. In step 1, paste https://mcp.getsalta.app into the connector settings. In step 2, approve once. Claude Code installs the server with two commands, and Cursor and Codex CLI take the same URL. PulseMCP and Glama crawl the Official Registry automatically, so a single publish propagates outward. In practice, one approval is the whole setup, and every write still waits for confirmation.

Why does MCP matter for AI assistants?

MCP is an open standard which matters because an assistant stops talking about your work and starts doing the work, without a custom integration for every app. Before the Model Context Protocol, connecting an AI to a real tool meant bespoke glue code. Such code worked in a single client and broke whenever either side changed. According to Anthropic's published specification, MCP makes the connection standard and reusable in both directions. First, build one server and every MCP client uses the server. Second, add one client and the client reaches every MCP server. Anthropic open-sourced the specification in late 2024, which is why adoption crossed vendors so quickly. For example, the same server answers Claude, Cursor, Codex CLI and ChatGPT in developer mode. In practice, your assistant stops being a place you copy things out of, and becomes something writing into the tools you already use, with your approval on every change.

How does Salta compare to the Todoist MCP?

The Todoist MCP server and Salta's MCP server both let an AI assistant act inside a real task app, and the two servers expose fundamentally different products. Todoist ships an official MCP server for a manual to-do list you build and maintain yourself — projects, labels, due dates, all designed by you. Notion ships an official MCP server for a workspace. Salta is an AI life planner, so the 18 MCP tools cover tasks, goals, and the sections a goal breaks into. For example, an assistant drafts an entire plan inside Salta, rather than filing items into a structure you already built by hand. Additionally, all three servers connect the same way, by pasting a server URL into a client, and all three keep data inside your own account. In practice, the difference is what the server is a server for — a list, a workspace, or a plan drafted with Claude and kept in Salta.

Questions

MCP, answered

What is MCP (Model Context Protocol)?

MCP, the Model Context Protocol, is an open standard for connecting AI assistants to external apps and data, so an assistant takes real actions instead of only describing them. Anthropic introduced and open-sourced MCP in late 2024 and published the full specification. MCP standardizes how AI clients — Claude, Claude Desktop, ChatGPT in developer mode, Cursor, Codex CLI — talk to MCP servers, which are the apps themselves. Before MCP, every AI-to-app connection was written from scratch and worked in exactly one client. Because the protocol is shared, one server now reaches every client speaking MCP, from Claude Desktop through Cursor and Codex CLI. Additionally, according to Anthropic's specification, a server advertises its own tools, so a client discovers what the server can do without manual configuration. In practice, MCP turns an assistant which describes your work into one which performs your work, with your approval on every change.

What does 'MCP server' mean?

An MCP server is an application which exposes data and actions to AI assistants through the Model Context Protocol. The client connects once, then reads information and performs real operations — creating, updating and deleting records in a live account, not simulating the work. For example, a file server lets an assistant read documents, and a calendar server lets an assistant create events. Salta's MCP server, at https://mcp.getsalta.app, publishes 18 tools covering pure CRUD over tasks and goals. Each tool is scoped to a single account and carries a risk hint, so the client knows before calling whether an action reads or writes. Additionally, no intelligence lives in the server: the assistant does the reasoning, calls the tools, and the user approves whatever lands. In practice, you connect with Claude, Claude Desktop, Cursor or Codex CLI, and the same 18 tools answer every one of them identically.

What are the Claude MCP servers, and how do I connect one?

Claude MCP servers are any MCP servers connected to Claude, Claude Desktop or Claude Code, and the same servers also work with Cursor, Codex CLI and ChatGPT's developer mode, because the protocol is shared rather than vendor-specific. Connecting a server takes two steps. In step 1, paste the server's URL into the client's connector settings. In step 2, approve the connection once. For Salta, the URL is https://mcp.getsalta.app, and Claude Code installs the server with two commands. Salta authenticates with OAuth 2.1, so you sign in to a real account rather than pasting an API key into a config file. Additionally, according to Anthropic's specification, a client discovers the server's tools automatically, so no manual setup follows the first approval. In practice, one approval is the whole installation, and every write still waits for confirmation before saving.

Who created MCP and is it open?

MCP was created and open-sourced by Anthropic, the maker of Claude, in late 2024. MCP is an open standard with a public specification, so any AI client or any application can implement the protocol without permission and without a commercial agreement. Openness explains the spread: a server built once is reachable from Claude, Cursor, Codex CLI and ChatGPT's developer mode, and a client adding MCP support immediately reaches every published server. Additionally, public directories index those servers — the Official MCP Registry at registry.modelcontextprotocol.io, plus Smithery, Glama, PulseMCP and mcp.so. For example, Salta is published on the Official Registry as app.getsalta/salta and scores 98/100 on Smithery's quality scan. Anthropic maintains the specification in the open, and any developer can publish a server. In practice, MCP is a protocol rather than a product, tied to no single vendor, and equally usable with Claude or any other compatible client.

How does Salta compare to the Todoist MCP?

The Todoist MCP server and Salta's MCP server both let an AI assistant act inside a real task app, and the two servers expose fundamentally different products. Todoist ships an official MCP server for a manual to-do list you build and maintain yourself: projects, labels, due dates. Notion ships an official MCP server for a workspace. Salta is an AI life planner, so the 18 MCP tools cover tasks, goals, and the sections a goal breaks into. For example, an assistant can draft an entire plan inside Salta, rather than filing items into a structure you already designed. Additionally, all three servers connect the same way, by pasting a server URL into a client, and all three keep data inside your own account. In practice, the difference is what the server is a server for: a list, a workspace, or a plan drafted with Claude and kept in Salta.