AI assistants (MCP)
The server also speaks the Model Context Protocol at /mcp
(Streamable HTTP, same bearer token), so an AI client can list, add, edit, schedule, complete and
delete tasks. It sits outside /api/v1 because MCP clients are configured with one URL.
Claude Code:
claude mcp add --transport http yarukoto https://todo.example.com/mcp \ --header "Authorization: Bearer $YARUKOTO_TOKEN"Claude Desktop, through the mcp-remote bridge in claude_desktop_config.json:
{ "mcpServers": { "yarukoto": { "command": "npx", "args": ["mcp-remote", "https://todo.example.com/mcp", "--header", "Authorization: Bearer YOUR_TOKEN"] } }}Tools: list_lists, list_tasks, get_task, create_task (takes quick-add text), update_task,
schedule_task, complete_task, delete_task (to Trash) and restore_task. Relative dates
resolve in YARUKOTO_TZ, and any call can pass its own timeZone.
The token grants full access, so the same caveat as the app applies: put the server behind HTTPS before pointing a client at it from outside your network. claude.ai’s custom connectors expect OAuth, which the server doesn’t offer yet.
