Docs / Concepts
What is MCP?
The protocol that connects your agent to Cornilius — explained plainly.
The protocol
MCP — Model Context Protocol — is an open standard that lets an AI agent call external tools over HTTP. An MCP server publishes a set of tools; the agent discovers them and calls them when a conversation needs them. The protocol is open and vendor-neutral: agents from different vendors can connect to the same server.
The server and the client
Cornilius is an MCP server. Your agent — Claude.ai, or any MCP-capable client — is the MCP client. When you ask an analytics question, the agent decides to call a Cornilius tool, and Cornilius answers from your workspace: your playbooks, your definitions, your recorded findings.
Your raw data sources are not connected to Cornilius. The agent queries those directly, the way it already does — Cornilius holds the method and the memory, not the data.
Your MCP key
An API key — the crn_ credential from the Connection page — identifies your workspace. Every call made with it reads and writes your data only; workspaces are isolated from each other. The key is shown once at creation, can be rotated with a grace window, and can be revoked at any time.
Claude Code and most other clients use the key. Claude.ai does not: it authenticates through a browser approval screen instead. Both paths end at the same place — calls scoped to your workspace.