Connectors (MCP)

X6 can use tools from Model Context Protocol (MCP) servers: course catalogs, calendars, databases, or anything else that offers an MCP server. Each server’s tools appear to X6 as mcp__<server>__<tool> and follow the same approval rules as X6’s own tools.

Adding a server

Servers are listed in an mcp.json file in X6’s data folder. Each one is either:

• a remote server (url, using Streamable HTTP), or
• a local command (command, using stdio).

Each server can also have headers or environment variables (secrets are written as references to the Keychain, never as plain text), an optional allow list of tool names, and an approval rule. Restart X6 after changing the file.

Approval rules

Rule	Behavior
auto	Read-only tools run freely; tools that change things ask.
always	Every tool from this server asks.
strict	Every tool asks, and Auto-approve never applies.

Servers that need sign-in

A remote server can sign in with standard MCP OAuth instead of a fixed key.

1. Open Settings → Connectors. Each server shows Ready, Needs sign-in, or Error.
2. Click Connect on a server that needs sign-in. The service’s sign-in page opens in a new tab.
3. Sign in. The tab says “Connected” once X6 has the tokens and has started the server.

X6 refreshes tokens before they expire. Disconnect deletes the sign-in and stops the server.

Where sign-ins are kept

Tokens are stored only in the operator’s Keychain: never in files, logs, the page, or what the model sees. Only a paired browser can start a sign-in, and each sign-in link works once, for ten minutes.

Safety

• A server that needs sign-in is skipped at startup, and X6 keeps running.
• Requests that carry a secret never follow redirects, so a server can’t forward the secret to another address.