MCP server settings gain connectorId and keep the configured command

What

An MCP server is an add-on that gives Claude extra tools. Its settings and handling changed in several ways:

• connectorId is a new, internal field for an MCP server that names the claude.ai connector the server proxies. The settings parser accepts a non-empty string; a non-string or empty value is reported in the list of ignored settings.
• For SDK MCP servers (servers supplied by a program that drives Claude Code through the SDK), connectorId applies when the server is first registered, like timeout and disableAutoBackground. A different connectorId sent again through initialize or setMcpServers is ignored, and the warning now names it.
• For sdk-type configs, connectorId is now removed along with tools from the server result.
• configuredCommand: when an MCP server config goes through environment-variable expansion, stdio servers (servers started as a local program) now keep their original command alongside the expanded one, and the server result reports it when it is found in the server's mcp_config.
• mcp_call and the permission prompt tool now refuse tools of certain servers. For example, a permission prompt tool on the MCP server remote-browser is denied.
• In one place the HTTP agent options are now built with carriesOurToken set to false.
• The copy picker gained new footer styling and a header label, and the settings panel has a tweaked layout for its search line.

Why

Program authors using the SDK should know that connectorId cannot be changed by re-sending initialize. The mcp_call change means a server such as remote-browser can no longer act as the tool that answers its own permission prompts.