feat(mcp): MCP server for AI agents, agent scopes on API keys, switches in settings

MindSift now speaks MCP (Streamable HTTP, stateless JSON-RPC) at POST /api/v1/mcp so
Claude Code, Codex, Claude Desktop, Cursor and custom agents can read, create and
complete tasks and keep shared lists. API keys get the scopes tasks:read/write and
lists:read/write; tools/list only shows what the key may call and every tool checks
its scope. Task and list write paths are shared helpers (updateTask, createTask,
addListItem, setListItemDone) used by REST and MCP alike.

Settings: scope selection and the SectionPicker (import, restore) use ToggleSwitch
rows instead of checkbox-like buttons; new MCP section with connection snippets for
Claude Code, Codex and JSON configs. Docs, README, help and changelog updated; v0.28.0.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

frontend/src/lib/types.ts

@@ -592,8 +592,15 @@ export interface UserTheme {
   updatedAt: string;
 }
 
-/** What an API key may deliver; keys never read data. */
-export type ApiKeyScope = 'webhook:triage' | 'webhook:task' | 'webhook:list';
+/** What an API key may do: webhook:* delivers data, the agent scopes read and change tasks/lists over MCP. */
+export type ApiKeyScope =
+  | 'webhook:triage'
+  | 'webhook:task'
+  | 'webhook:list'
+  | 'tasks:read'
+  | 'tasks:write'
+  | 'lists:read'
+  | 'lists:write';
 
 export interface APIKey {
   id: string;