The agent data glossary
An MCP setup exposes a tool; it does not grant data permission. A validated recipe is not an executed workflow. A skill is an instruction file, not a guarantee. These definitions keep the boundary visible.
MCP and clients
MCP
Model Context Protocol, a standard way for an AI client to discover and call external tools and data sources. More on MCP
MCP server
The process or remote endpoint that exposes tools to an MCP client. Apify's MCP server can expose one Actor or a wider set of tools. More on MCP server
MCP client
The application that connects to an MCP server, such as Claude Desktop, Claude Code, Cursor or a ChatGPT connector. More on MCP client
stdio
A local transport where the client starts the server process and communicates over standard input and output. The Claude Desktop, Claude Code and Cursor configs on this site use it. More on stdio
Remote HTTP
A transport where the client connects to a hosted URL. ChatGPT custom connectors use a remote endpoint rather than a local JSON process. More on Remote HTTP
OAuth
An authorization flow that lets a client obtain access without a pasted long-lived token. The remote MCP route can use the client's authentication flow. More on OAuth
Connector
A client-level integration that adds external tools to a chat or agent. A connector may use a remote URL, a local process or a marketplace app. More on Connector
Actors and files
Actor
A packaged program on Apify that accepts structured input, runs on the platform and writes output to a dataset. More on Actor
Dataset
The structured output of an Actor run, available as JSON, CSV or Excel. A dataset can contain nested fields and arrays. More on Dataset
Recipe
An importable n8n workflow that connects a catalogue Actor to a destination such as Sheets, Slack, Notion or email. More on Recipe
Skill
A packaged instruction file, with a script and example input, that tells an agent when and how to perform a bounded data task. More on Skill
n8n workflow
A node-based automation. The recipes on this site are validated for JSON structure, known nodes, connections and required parameters, not executed. More on n8n workflow
Credential
A stored authentication value in n8n, such as an Apify token or a destination API key. The recipe files contain placeholders, not credentials. More on Credential
Runs and access
Run
One execution of an Actor with an input. A run produces a dataset and public run statistics. More on Run
Run-sync
An API pattern that starts an Actor run and waits for the dataset in the same request. It suits small jobs; long runs may time out. More on Run-sync
API token
A secret used to authenticate an Apify API call. Keep it in the client or automation secret store; never commit it to a shared repository. More on API token
Input schema
The published fields an Actor accepts, including required values and types. The source pages show the input used by the checked file. More on Input schema
File coverage
The share of catalogue Actors that have at least one recipe, MCP config or skill on this site. The catalogue page shows the gaps as well as the ready routes. More on File coverage
Questions
What is the difference between an MCP server and an Actor?
The Actor does the data work. The MCP server exposes it to an AI client. A local MCP config starts a server process; a remote connector points at a hosted endpoint.
Does a validated recipe run the Actor?
No. The recipe is validated for n8n structure, connections and Actor input schema. It is not executed. You still need an Apify token and the destination credential.
Where should I store an API token?
In the client or automation secret store, not in a repository, screenshot or public issue. The checked configs contain placeholders.