Skip to main content

MCP Connectors

An MCP connector (Model Context Protocol) gives an AI agent access to an external tool during its analysis — calculation, dictionary, encyclopedic lookup, calendar, or any other service exposed via this protocol.

Built-in connectors​

Four connectors are pre-installed and protected (their configuration cannot be edited or deleted — only their enabled state remains changeable):

  • math — calculation tools.
  • wikipedia — encyclopedic lookup.
  • hunspell — spellchecking; always run deterministically before the LLM call on agents that enable it, with its findings injected back even if the AI doesn't pick them up on its own.
  • calendar — date/calendar-related tools.

Adding a custom connector​

From Admin → MCP Connectors, two transport types:

  • stdio — launches a local command (executable/script) accessible from the server; requires providing the command and its arguments.
  • http — connects to a remote MCP server via a URL, with customizable HTTP headers (for authentication, for instance).

Usage​

A connector (built-in or custom) must be explicitly enabled to be usable, then attached to one or more agents — either from the agent's full configuration, or directly via the dedicated checkbox in the AI Agent node panel of the pipeline editor.